Metabolomic studies in the inborn error of metabolism alkaptonuria reveal new biotransformations in tyrosine metabolism

Alkaptonuria (AKU) is an inherited disorder of tyrosine metabolism caused by lack of active enzyme homogentisate 1,2-dioxygenase (HGD). The primary consequence of HGD deficiency is increased circulating homogentisic acid (HGA), the main agent in the pathology of AKU disease. Here we report the first metabolomic analysis of AKU homozygous Hgd knockout (Hgd−/−) mice to model the wider metabolic effects of Hgd deletion and the implication for AKU in humans. Untargeted metabolic profiling was performed on urine from Hgd−/− AKU (n = 15) and Hgd+/− non-AKU control (n = 14) mice by liquid chromatography high-resolution time-of-flight mass spectrometry (Experiment 1). The metabolites showing alteration in Hgd−/− were further investigated in AKU mice (n = 18) and patients from the UK National AKU Centre (n = 25) at baseline and after treatment with the HGA-lowering agent nitisinone (Experiment 2). A metabolic flux experiment was carried out after administration of 13C-labelled HGA to Hgd−/−(n = 4) and Hgd+/−(n = 4) mice (Experiment 3) to confirm direct association with HGA. Hgd−/− mice showed the expected increase in HGA, together with unexpected alterations in tyrosine, purine and TCA-cycle pathways. Metabolites with the greatest abundance increases in Hgd−/− were HGA and previously unreported sulfate and glucuronide HGA conjugates, these were decreased in mice and patients on nitisinone and shown to be products from HGA by the 13C-labelled HGA tracer. Our findings reveal that increased HGA in AKU undergoes further metabolism by mainly phase II biotransformations. The data advance our understanding of overall tyrosine metabolism, demonstrating how specific metabolic conditions can elucidate hitherto undiscovered pathways in biochemistry and metabolism.


Introduction
Alkaptonuria (AKU) is a rare disorder of tyrosine metabolism caused by congenital lack of activity of the enzyme homogentisate 1,2-dioxygenase HGD (E.C.1.12.11.5). 1 The biochemical consequence of HGD deficiency is increased homogentisic acid (HGA) in the circulation, the pathognomonic sign of the disease and central to its pathophysiological features. 2 HGA has a high affinity for collagenous tissues, where its deposition produces striking pigmentation, 3 a process called ochronosis. Cartilage of load-bearing joints is particularly susceptible to ochronosis. Presence of HGA-derived pigment in these joints alters the physicochemical properties of cartilage that support normal transmission of load and results in an inevitable and severe early-onset osteoarthropathy. 4,5 Metabolomics has emerged as an invaluable approach for studying AKU. In our laboratory we have developed a targeted approach with specific mass spectrometric assays as an aid for diagnosis and monitoring of AKU. 6,7 These assays offer precise quantification of tyrosine pathway metabolites including HGA in serum and urine. More recently, a strategy has been developed for profiling and chemical identification of up to 619 related and more general metabolites by high-resolution accurate mass and retention time using liquid chromatography quadrupole time-of-flight mass spectrometry (LC-QTOF-MS). 8 Application of this technique to AKU serum 9 and urine 8 enabled the discovery of previously unknown metabolite and metabolic pathway alterations following treatment with the HGA-lowering agent nitisinone. Metabolic profiling therefore has potential in AKU as both a phenotyping and biomarker discovery tool. However, to our knowledge, untreated AKU has not been compared with non-AKU at the metabolome level before.
The investigations into such possible changes in the metabolome as a result of AKU will be greatly facilitated by use of an animal model of AKU developed in our laboratory by homozygous knockout of the Hgd gene (Hgd À/À ) in mice. The genetically altered Hgd À/À mouse recapitulates human AKU, with elevated plasma and urine HGA, development of ochronosis and its inhibition by nitisinone. 10,11 It was important to establish the metabolic profile of this model, so we first compared the urinary profiles of AKU homozygous Hgd knockout mice (Hgd À/À ) with non-AKU heterozygous knockout (Hgd þ/À ) control mice (Experiment 1). Secondly, we looked to confirm any AKU-related metabolite differences by assessing whether the direction of alteration was reversed while on the HGA-lowering drug nitisinone in Hgd À/À mice or patients (Experiment 2). Thirdly, a metabolic flux experiment was carried out (Experiment 3), in which mice were injected with stable isotope 13 C-labelled HGA to ascertain whether metabolites increased in Hgd À/À were directly derived from HGA.
The overall aim of these experiments ( Fig. 1) was to investigate in a controlled study the metabolic markers of disease in AKU and if metabolite alterations could elucidate possible novel disease mechanisms and thereby improve current strategies for monitoring the disease process and subsequent treatment response in patients with AKU.

Animal housing and husbandry
All mice were housed in the University of Liverpool's Biomedical Services Unit under pathogen-free conditions, in cages of up to five mice, with 12-hr light/dark cycle, and food and water available ad libitum. Mice were drug/testnaïve at baseline in each experiment.

Urine collection and sample preparation for metabolomics (Experiment 1)
For metabolomic analysis of the targeted Hgd knockout phenotype, 10 urine was collected from 15 Hgd À/À (mean age AE SD 12.8 AE 0.1 weeks) and 14 Hgd þ/À (mean age 11.6 AE 0.3 weeks) male C57BL/6 mice. Drinking water was supplied ad libitum and the mouse urine was collected on a single-collection basis onto plastic wrap, pipetted into sample tubes and stored at À80 C. Increased urinary HGA was expected to be the most marked metabolic alteration in Hgd À/À , but we were also interested to study the wider, potentially more subtle metabolic alteration both within the tyrosine pathway and in non-directly associated pathways. The number of mice studied was therefore considered appropriate to sufficiently power this experiment, given that mouse urine collection is a non-invasive procedure.
Pooled samples were created in each profiling experiment for quality control (QC) purposes. For Hgd À/À and Hgd þ/À groups a separate representative pool was created by pooling 20 mL of each individual urine sample. An overall pool was also created for each experiment by pooling equal proportions of the above group pools. Analysis of individual and pooled samples was performed following dilution of 1:9 urine:deionised water as previously described. 8 Investigating the effect of nitisinone on metabolites showing alteration in Hgd À/À mice (Experiment 2) The effect of nitisinone treatment on urinary metabolites altered in Hgd À/À mice (Fig. 1, Experiment 1) was studied in the data from a previous profiling experiment described by Norman et al. 8 These data were from urine from 18 BALB/c Hgd À/À mice (mean age 27 AE 12 weeks, 9 female, 9 male) with Hgd disruption by ENU mutagenesis 11 and from 25 patients attending the UK National Alkaptonuria Centre (NAC; mean age 51 AE 15 years, 13 female, 12 male). The disease Figure 1 Schematic overview of the overall study design, incorporating Experiments 1e3. In Experiment 1, urine was collected from Hgd À/À and Hgd þ/À mice and profiled by LC-QTOF-MS. Targeted and non-targeted feature extraction was performed on the data in parallel and subsequent unpaired t-tests were employed to identify differentially abundant compounds between Hgd À/À and Hgd þ/À . These compounds were then further investigated in LC-QTOF-MS data from two additional datasets; a previously published study examining the effect of nitisinone on the urine metabolome of Hgd À/À BALB/c mice and patients with AKU 8 (Experiment 2) and a plasma flux analysis using a 13 C 6 labelled HGA tracer (Experiment 3). phenotypes of Hgd À/À mice from targeted knockout and ENU mutagenesis models are identical. 10 Data were from urine collected at baseline, then on nitisinone at one week in mice (supplied ad libitum in drinking water at 4 mg/L) and at 24 months in patients (2 mg daily dose). These datasets were acquired under identical LC-QTOF-MS analytical conditions to those employed in the present study.
Design of in vivo metabolic flux experiment and sample collection (Experiment 3) Eight C57BL/6 mice were studied in the HGA metabolic flux experiment; four Hgd À/À (mean age 56 AE 2.3 weeks, 1 female, 3 male) and four Hgd þ/À (mean age 58 AE 0 weeks, 4 female). A 1.96 mg/mL 13 C 6 HGA tracer solution was prepared in sterile saline and injected into the tail vein. Injection volume was adjusted for each mouse to achieve a final blood concentration of 1 mmol/L, assuming a total blood volume of 75 mL/kg. 12 Venous tail bleed samples were then taken at 2, 5, 10, 20, 40 and 60 min postinjection, keeping sampling volumes within LASA guidelines. 12 Mice were kept anaesthetised with isoflurane throughout the experiment. Blood was collected into Microvette 300 mL lithium heparin capillary tubes (Sarstedt, Nümbrecht, Germany) and centrifuged at 1500Âg for 10 min. Plasma supernatant was removed and stored at À80 C prior to analysis. Individual plasma samples were analysed following 1:9 plasma:deionised water.

LC-QTOF-MS analyses
Analysis of plasma and urine was performed on an Agilent 1290 Infinity HPLC coupled to an Agilent 6550 QTOF-MS equipped with a dual AJS electrospray ionization source (Agilent, Cheadle, UK). Reversed-phase LC was performed on an Atlantis dC18 column (3 Â 100mm, 3 mm, Waters, Manchester, UK) maintained at 60 C. Mobile phase composition was (A) water and (B) methanol, both with 5 mmol/L ammonium formate and 0.1% formic acid. The elution gradient began at 5% B 0e1 min and increased linearly to 100% B by 12 min, held at 100% B until 14 min, then at 5% B for a further 5 min. MS data acquisition was performed in positive and negative ionisation polarity with mass range 50e1700 in 2 GHz mode with acquisition rate at 3 spectra/ second. Sample injection volume was 2 mL, and the autosampler compartment was maintained at 4 C. Additional data acquisition parameters are detailed in Appendix 1.
Data-dependent tandem mass spectrometry (MS/MS) was performed on pooled urine samples, with compound hits from Experiment 1 as [MþH] þ and [MÀH]accurate mass precursor ion targets; no more than six compound targets per injection. Fixed collision energies of 10, 20 and 40 V were applied. Acquisition rates were 6 spectra/second in MS and 4 spectra/second in MS/MS.

Design of LC-QTOF-MS profiling analyses
Samples from Experiments 1 (Hgd À/À versus Hgd þ/À ) and 3 ( 13 C 6 HGA metabolic flux analysis) were analysed in separate batches, each comprising negative then positive polarity. The analytical sequence of each profiling batch was designed according to published guidance 13 and following the procedure described previously by Norman et al. 8

Data processing and statistical analyses
Mining of metabolite features in raw data was performed using two parallel approaches (Fig. 1, Experiment 1). A targeted approach was used to extract signals matching a 466-compound AMRT database previously generated in our laboratory from IROA Technology MS metabolite library of standards, accessible via https://doi.org/10.6084/m9. figshare.c.4378235.v2, 8 or a compound database comprising accurate masses of additional compounds with potential relevance to AKU. A complementary non-targeted approach was used to extract unknown metabolites.
Targeted feature extraction Targeted feature extraction was performed in Profinder (build 08.00, Agilent) using the chemical formulae of compounds from the AMRT database described above. Extraction parameters were accurate mass match window AE10 ppm and retention time (RT) window AE0.3 min. Allowed ion species were: H þ , Na þ , and NH 4 þ in positive polarity, and H À and CHO 2 in negative polarity. Charge state range was 1e2, and dimers were allowed. 'Find by formula' filters were: score >60 in at least 60% of samples in at least one sample group (samples were grouped by Hgd À/À or Hgd þ/À and pre-or on nitisinone).
Seventy-five additional metabolites of potential interest in AKU or from wider tyrosine metabolism were appended to the database for targeted extraction. Forty-three were from the following pathway databases available from Pathways to PCDL (build 07.00, Agilent): 'citrate degradation', 'noradrenaline and adrenaline degradation' and the 'superpathway of phenylalanine, tyrosine and tryptophan biosynthesis' (Appendix 2). Six metabolites were added as they were predicted to show potential alteration in Hgd À/À due to association with tyrosine conjugation (acetyl-Ltyrosine, and g-glutamyl-tyrosine) based on a previous publication 8 or association with ochronotic pigment derived from HGA (2,5-dihydroxybenzaldehyde, benzoquinoneacetic acid, hipposudoric acid and norhipposudoric acid). Twenty-six metabolites were from a list of potential biotransformation products directly derived from HGA and compiled using the Biotransformation Mass Defects application (Agilent; Appendix 3). This tool provides a list of potential metabolic biotransformation products covering both phase I and II metabolism for a given compound based on empirical formula. The data were mined for these additional metabolites with putative identification by accurate mass (AE5 ppm) only.
Non-targeted feature extraction Non-targeted extraction was performed by recursive feature extraction in Profinder (build 08.00). Extraction parameters are detailed in Appendix 4.
Isotopologue feature extraction on data from 13 C 6 HGA metabolic flux analysis Data from Experiment 3 ( Fig. 1) were mined using batch isotopologue extraction in Profinder (build 08.00). Here, compounds that showed significant differences between Hgd À/À and Hgd þ/À mice (Experiment 1) were investigated for potential association with the 13 C 6 HGA tracer by examining the relative abundances of their Mþ0 to Mþ6 isotopologues. Extraction was performed with accurate mass and RT match windows of AE5 ppm and AE0.3 min respectively against an AMRT database consisting only of these compound targets.
Data QC and statistical analyses First, initial QC was performed on all datasets in Profinder by manual curation of the dataset to remove visually low quality peaks and to correct integration issues across the dataset where appropriate.
Urine profiling datasets were then exported from Profinder as .csv files and imported into Mass Profiler Professional (build 14.5, Agilent) for additional QC and subsequent statistical analyses. First, creatinine peak area from this analysis was used as an external scalar for each mouse sample to account for differences in urine concentration, as described previously. 8 Additional QC was performed using data from pooled samples, which were interspersed throughout each analytical sequence. Compounds were retained if a) observed in 100% of replicate injections for at least one sample group pool, and b) peak area coefficient of variation (CV) remained <25% across replicate injections for each sample group pool. Statistically significant compounds were then identified in each dataset by t-tests; two sample t-tests for Hgd À/À versus Hgd þ/À comparisons, and paired t-tests for pre-versus on nitisinone comparisons. Multiple testing correction was performed using Benjamini-Hochberg false discovery rate (FDR) adjustment. Fold changes (FC's) were log 2 -transformed and based on peak area. FC cut-off was not applied for compounds with FDR-adjusted P < 0.05, in order to retain compounds that showed consistent although relatively low magnitude abundance differences between comparison groups. Principal component analysis (PCA) was performed on each filtered dataset using four-component models.

In vivo metabolic flux data analysis
The results from isotopologue extraction on plasma 13 C 6 HGA metabolic flux data were reviewed visually in Profinder for clear evidence of an isotope label likely to be derived from HGA. This data review was performed for compound matches individually using the predicted number of 13 C atoms derived from the HGA tracer based on chemical structure.

MS/MS data analysis
Raw MS/MS data files were processed in MassHunter Qualitative Analysis Workflows (Build 08.00), using the targeted MS/MS compound discovery algorithm. For compounds from the in-house AMRT database, MS/MS spectral library matching was performed against the METLIN metabolite PCDL accurate mass library (build 07.00), which contains over 30,000 compound entries, of which 1804 contain experimental MS/MS spectral data. For the HGA-derived compounds, for which no known chemical standards or compound library entries currently exist, structure identifications were performed using Agilent Molecular Structure Correlator (version B.07.00, build 31). This approach involves matching between accurate mass MS/MS fragment ions obtained from collision-induced dissociation MS with one or more proposed molecular structures based on a systematic bond disconnection approach. 14

Study approval
All animal work was carried out in accordance with UK Home Office Guidelines under the Animals (Scientific Procedures) Act, 1986 and with institutional approval.
Metabolomic analyses on patient samples was approved by the Royal Liverpool and Broadgreen University Hospital Trust's Audit Committee (audit no. ACO3836) and was part of the diagnostic service to patients attending the UK National Alkaptonuria Centre in Liverpool.

Results
PCA showed clear separation in principal component 1 between the urine profiles of Hgd À/À and Hgd þ/À mice from targeted ( Fig. 2A, B) and non-targeted feature extraction. Results from targeted and non-targeted extraction are presented separately in the following sections (number of compounds obtained in each extraction method summarised in Appendix 5). Data from all animals in each experiment were included in the analysis.

Targeted feature extraction
Targeted feature extraction was performed to search for metabolites based on AMRT (accurate mass AE10 ppm, RT AE0.3 min) or accurate mass (AE5 ppm) alone. 27/250 and 15/243 metabolites showed abundance differences (FDRadjusted P < 0.05) between Hgd À/À and Hgd þ/À in negative and positive polarity, respectively. Table 1 shows the altered metabolites ranked by FC. PCA loadings plots (Fig. 2C, D) and volcano plots (Fig. 2E, F) show that the greatest differences between Hgd À/À and Hgd þ/À urine were in metabolites associated with HGA in negative and positive polarity. HGA and six predicted HGA biotransformation products were markedly elevated in Hgd À/À . Interestingly, HGA-sulfate (FC Z 9.3, P < 0.0001) showed a greater FC increase than HGA (FC Z 7.9, P < 0.0001). Other HGA products increased with FC > 1.5 and P < 0.0001 were HGA-glucuronide, HGA-hydroxylsulfate, HGA-N-acetylcysteine and HGA-hydrate. Acetyl-HGA was elevated in Hgd À/À (FC Z 2.3, P < 0.0001), but did not pass QC filtering by CV <25% across replicate injections of pooled QC samples; a decrease in signal across the run indicated a stability issue (despite the auto-sampler being maintained at 4 C).
MS/MS fragmentation data were acquired to confirm the compound identifications of the significantly altered metabolites. For the HGA-derived metabolites, the proposed structures (Fig. 4) were imported into the in-house AMRT compound library as .mol files for matching of experimental MS/MS spectra against in silico fragmentation predictions using Molecular Structure Correlator (MSC). MSC match scores >85% were obtained for HGA-glucuronide (91.2%), HGA-sulfate (86.7%) and HGA-hydroxylsulfate (85.9%), strongly supporting the chemical structures proposed for these compounds. Intermediate scores were obtained for the proposed structure for acetyl-HGA (54.3%) and hydroxymethyl-HGA (53.9%). Lower MSC scores were obtained for the structures proposed for HGA-N-acetylcysteine (32.4%) and HGA-hydrate (20.3%). The MSC score obtained for each of these HGA-derived structures was either the only compound structure match or was greater than the highest scoring match from the MassHunter METLIN metabolite library. For the significantly altered metabolites that were not HGA biotransformation products, 15 of the AMRT-matched compounds were confirmed by MS/MS Figure 2 Clear differences between the urine metabolomes of Hgd À/À and Hgd þ/À mice. (AeD) PCA on data from targeted feature extraction, with PCA plots showing separation between Hgd À/À and Hgd þ/À mice by component 1 in A, negative, and B, positive ionisation polarities. Lower plots show the corresponding PCA loadings of metabolites on components 1 and 2 in C, negative, and D, positive polarity. (E, F) Volcano plots illustrating selection of statistically significant urinary metabolites between Hgd À/À and Hgd þ/À mice based on p-value and fold change. (E) negative polarity; (F) positive polarity. Compounds with P < 0.05 (Benjamini-Hochberg FDR adjusted) and log 2 fold change >1.5 are labelled, with red and blue indicating increased and decreased abundance, respectively, in Hgd À/À . Turquoise indicates adjusted P < 0.05 but log 2 fold change <1.5. Bold text indicates that the increase observed in Hgd À/À was confirmed in mouse plasma following injection with 13

Non-targeted feature extraction
Non-targeted feature extraction yielded 359 and 213 compounds post-QC in negative and positive polarity respectively; mass range Z 54e3108 Da and RT range Z 1.1e11.5 min. Comparison of Hgd À/À and Hgd þ/À revealed compounds with clear abundance differences (P < 0.05, log 2 FC > 1.5); 9 in negative polarity (6 increased in Hgd À/À , 3 decreased in Hgd À/À ) and 2 in positive polarity (both increased in Hgd À/À ). The mass range of these significant compounds was 242e507 Da, and RT range was 1.7e9.6 min ( Table 1). Searching of the MassHunter METLIN metabolite PCDL accurate mass library (build 07.00) yielded no matches for these compounds based on library MS/MS spectral matching or accurate mass (<5 ppm) alone.
The effect of nitisinone treatment on metabolites altered in Hgd À/À mice To investigate the effect of nitisinone treatment on the metabolites shown to be altered here in Hgd À/À , they were searched in the data from previous mouse and human urine profiling experiments (Experiment 2; Fig. 1). 8 Eight compounds from targeted feature extraction that were altered in Hgd À/À versus Hgd þ/À mice were significantly altered in the opposite direction in urine from both Hgd À/À mice and patients with AKU on nitisinone (based on peak area pre-versus on nitisinone, Benjamini-Hochberg FDR-adjusted P < 0.05; Table 2); seven decreased, and one increased. The seven decreased metabolites were HGA, the HGA biotransformation products HGA-sulfate, HGA-glucuronide, HGA-hydrate and hydroxymethyl-HGA, and also xanthosine and 3,5-cyclic-AMP. The increased metabolite was 3-methyl-glutaric acid. Interestingly, phydroxyphenylacetic acid and 4-hydroxybenzaldehyde were increased both in Hgd À/À versus Hgd þ/À and also on nitisinone in Hgd À/À mice and patients.
Nineteen compounds that were altered in Hgd À/À versus Hgd þ/À mice were significantly altered in the opposite direction on nitisinone in urine from either Hgd À/À mice (n Z 14; 9 decreased, 5 increased on nitisinone) or patients (n Z 5; 4 decreased, 1 increased on nitisinone) only ( Table  2). These compounds comprised the remaining HGA biotransformation products, which were all decreased on nitisinone. On nitisinone, acetyl-HGA was decreased in patients only, and HGA-hydroxyl-sulfate and HGA-N-acetylcysteine were decreased in mice only.

Confirmation of HGA biotransformation products by 13 C 6 HGA metabolic flux analysis (Experiment 3)
Data from isotopologue extraction were compared between plasma collected from the same mice across the time intervals available (2e60 min). The Mþ6 isotopologue was of particular interest as mice were injected with 13 C 6 -labelled HGA. A clear Mþ6 peak for HGA was observed over the sampling time course in plasma from Hgd À/À and Hgd þ/À mice, although the signal decreased from 10 to 20 min post-injection (Fig. 3). As indicated in Figure 2 (compounds in bold text), two HGA biotransformation products that were increased in Hgd À/À urine were observed with clear Mþ6 peaks over the time course; HGA-glucuronide and HGAsulfate. These data confirm that the compounds are derived from HGA. HGA-sulfate showed a similar time course profile to HGA; the native Mþ0 isotopologues were absent from Hgd þ/À plasma at all time points, in contrast to the Mþ6 isotopologue whose profile appeared to closely follow that of the HGA Mþ6 peak over the time course in both Hgd À/À and Hgd þ/À (Fig. 3). For HGA-glucuronide, the native Mþ0 isotopologue was also absent from Hgd þ/À plasma, but the Mþ6 isotopologue was only observed in Hgd À/À (Fig. 3), indicating that glucuronidation of the HGA tracer was evident only for Hgd À/À mice.

Discussion
The data reported form part of the first recorded metabolome-wide comparison obtained from untreated AKU versus non-AKU animal models and shows, for the first time, that the overproduced HGA in AKU undergoes predominantly phase II metabolic biotransformations. Phase II metabolism involves conjugation reactions to form sulfate, glucuronide, glutathione, mercapturic acid, amino acid, methyl and acetyl conjugates. 15,16 The predominant sulfation and glucuronidation biotransformations observed for HGA are also essential aspects of metabolism of other phenolic acids, bile acids, steroids and numerous other endogenous biochemicals. The phase I metabolism reactions of hydroxylation, oxidation, reduction and hydrolysis were relatively minor. Further, our data indicate that the biochemical consequences of HGD deficiency extend beyond tyrosine metabolism.

HGA biotransformation products from phases I and II metabolism
The clearest differences between the urine metabolomes of Hgd À/À versus Hgd þ/À mice were in HGA and seven previously unreported HGA-derived biotransformation products, increased in Hgd À/À . The decreased output in HGA-sulfate, HGA-glucuronide, HGA-hydrate, acetyl-HGA and hydroxymethyl-HGA for patients on treatment with nitisinone (Experiment 2) indicates that these HGA phase II biotransformations occur in human AKU. The detection of 13 C 6 -labelled forms of HGA-glucuronide and HGA-sulfate in plasma following 13 C 6 -HGA injection in mice (Experiment 3) confirm products derived from HGA. The 13 C 6 -HGA-sulfate was observed both in Hgd À/À and Hgd þ/À mice, following the profile of the 13 C 6 -HGA across the sampling time course. The 13 C 6 -HGA-glucuronide was observed only in Hgd À/À , suggesting prior upregulation of glucuronyltransferase activity was required to enhance HGA clearance in AKU.
The putative structures of these newly identified biotransformation compounds, together with the enzyme classes for their formation are shown in Figure 4, although the exact enzymes are not yet described. The proposed biotransformation compound structures are the closest known matches, based on correlation between the observed experimental MS/MS spectra and the accurate Table 1 Summary of urinary metabolites showing altered abundance in Hgd À/À mice.
Direction of alteration and log 2 fold change is indicated in Hgd À/L (relative to Hgd þ/L ); red and blue shading indicates increased and decreased abundance in Hgd À/L , respectively. p-values are false discovery rate adjusted. Where compounds were significantly different in positive and negative polarity, the result with the lowest fold change is provided. For compounds identified by accurate mass (AM) and retention time (RT), match criteria were accurate mass (AE10 ppm) and RT (AE0.3mins) against a database generated in-house from metabolite standards. Compound identifications based on accurate mass match alone were with a mass mass fragment ions predicted in silico from the 30,000þ metabolite structures in the Metlin and in-house compound libraries. In the absence of reference standards for these compounds, the MS/MS data confirm the structures proposed for HGA-glucuronide, HGA-sulfate and HGA hydroxylsulfate, with MSC scores >85%. To our knowledge, these are the first data on HGA metabolism not involving conversion to maleylacetoacetic acid in the traditional tyrosine degradation pathway (Fig. 5).
The markedly increased output of the HGA-glucuronide (FC Z 5) and HGA-sulfate (FC Z 9.3), as well as HGA (FC Z 7.9), in AKU Hgd À/À mice has implications for the regulation of plasma HGA and products. Despite efficient renal excretion of HGA-related compounds in AKU, 17 the circulating HGA remains elevated; mean serum HGA concentration is 30 mmol/L in untreated AKU 18 and <1.5 mmol/ L in healthy subjects. 19,20 The increased plasma HGA persists as the primary toxic agent in AKU; its oxidation produces ochronotic pigment, which becomes bound within the extracellular matrix of collagenous tissues. This process of damage is thought to occur via HGA oxidation to highly reactive benzoquinone and free radical intermediates. 5 These species are capable of inducing oxidative changes to proteins and lipids, 21,22 are known to self-perpetuate osteoarthropathy associated with ochronosis 5 and account for cases of acute fatal metabolic consequences reported in the literature. 23 The data reported here show that metabolism of HGA by phase II biotransformations, particularly glucuronidation and sulfation reactions, is a major route in an attempt to render the increased HGA chemically inert (detoxify), in effect to protect from consequences of increased benzoquinone production. The HGA-derived products observed must be included in metabolic profiling in AKU patients to understand the total production of HGA and how this impacts on predicting the amount of HGAderived pigment produced in vivo.
It is not clear whether the HGA biotransformations described are simply unmasked in AKU due to the markedly increased HGA, or whether they are actively and exclusively recruited in AKU for HGA detoxification. Nevertheless, identification of the specific enzymes that catalyse the biotransformations could inform future AKU therapeutic interventions aimed at enhancing HGA metabolism. Glucuronidation, for example, is quantitatively one of the most important phase II biotransformation reactions, performed by 15 UPD-glucuronosyltransferase enzymes for conjugation of a large number of exogenous and endogenous compounds in humans. 24,25 A number of agents, including naturally occurring dietary compounds, are known to be potent inducers of UPD-glucuronosyltransferases and other phase II enzymes. 26e28 There are encouraging examples of targeting specific enzymes of phase II metabolism in other conditions of endogenous compound accumulation, including UDPglucuronosyltransferase 1A1 in neonatal hyperbilirubinaemia 29 and sulfotransferases in oestrogendependent breast cancer. 30 The liver is the primary site of phase I and II metabolism, both by tissue mass and activity. 31 Extrahepatic metabolism in tissues such as the kidney and intestinal epithelium is also significant. 32 It is possible that site-specific enzyme activity explains the interesting finding that conjugation products from the 13 C 6 -labelled HGA were observed mainly for glucuronidation and sulfation. The other biotransformation products observed in urine were not detected in the plasma tracer experiment, despite their clear concomitant urinary reduction on nitisinone confirming direct association with HGA. In the metabolic flux experiment, HGA was administered intravenously, which is likely to favour window AE5 ppm. a MS/MS compound identification based on matching experimental spectra with in silico fragmentation data, using Molecular Structure Correlator (MSC) with score threshold >65%; all other MS/MS matches were against spectra from the Mass-Hunter METLIN metabolite PCDL accurate mass library (build 07.00). b Acetyl-HGA failed quality control filtering (CV >25% across replicate injections of QC pooled samples; due to suspected compound stability issue over analysis period).
formation of products from hepatic metabolism. It would be interesting therefore to give HGA by various routes of delivery and trace its metabolism in various tissues and biofluids to assess the metabolic activity contributed by different anatomical compartments. It is also worth noting that the glycine conjugate of HGA, a potential HGA biotransformation product, was not detected in Hgd À/À ; glycine conjugation is a known detoxification mechanism of a number of other aromatic acids. 33,34 The observation of marked elevations in five of the seven HGA-derived compounds which were initially discovered in mice, including the major biotransformation products from HGA sulfation and glucuronidation, in untreated patients with AKU (Experiment 2) supports use of the Hgd À/À mouse as an accurate model of human AKU biochemistry. Our preclinical studies of the Hgd À/À mouse have helped us to understand the metabolic and wider pathophysiological features of AKU in humans, the natural course of the disease and its response to treatment. The mice studied in Experiment 1 were male only in order to enable a highly controlled and powered examination of the metabolic impact of Hgd knockout. The inclusion of the previously published data from equal male and female Hgd À/À mice in Experiment 2 confirms that the metabolic alterations observed for AKU mice in Experiment 1 are not restricted to males. Further, in Experiment 3 we observed both native non-labelled and 13 C-labelled HGA-glucuronide and HGA-sulfate products for male and female Hgd À/À mice. As noted in our published discussion of the dataset featured in Experiment 2, the metabolomic difference between urine from male and female mice is largely attributable to the markedly increased urinary histamine in female mice, and no difference in metabolism of tyrosine or HGA. 8 Associated alteration to tyrosine, purine and TCA cycle metabolism in AKU This study showed for the first time that in untreated AKU there is alteration to tyrosine, purine and TCA cycle metabolites (Fig. 5). Previous metabolomic studies in AKU have focused solely on the impact of nitisinone on the metabolome. 8,9,35e37 Nitisinone reversibly inhibits hydroxyphenylpyruvic acid dioxygenase (HPPD; E.C. 1.13.11.27), the enzyme that produces HGA, and it is currently the most effective treatment for AKU. Nitisinone reduces plasma and urine HGA concentrations, 18,19,38e40 completely arrests ochronosis in AKU mice 11,41 and more recently was shown to decrease ochronosis and improve clinical signs of AKU in patients enrolled in SONIA 2, the 4-year phase 3 international randomised controlled trial of nitisinone in AKU (ClinicalTrials.gov identifier: NCT01916382). 42 We showed previously in serum and urine that nitisinone induced an extended network of metabolic alteration within tyrosine and neighbouring pathways, including tryptophan, purine and TCA cycle. 8,9 This alteration is a concern in AKU, and particularly in hereditary tyrosinaemia type-1, another inherited disorder of tyrosine metabolism, in which nitisinone treatment is essential from early infancy. 43 The increases in tyrosine metabolites reported here, excluding HGA, were unexpected in untreated AKU. Increases in metabolites upstream of HPPD previously reported in nitisinone-treated AKU were thought to be a Figure 4 Predicted structures of newly-identified HGA biotransformation products resulting from phase I and II metabolism. Predicted structures were the closest matches against the acquired experimental MS/MS data, based on scores obtained using Agilent Molecular Structure Correlator (MSC). The proposed sites for metabolism/conjugation were based on match scores obtained for a list of possible candidates using a combination of MSC and CFM-ID 3.0 53 in silico fragmentation modelling tools. Table 2 The effect of nitisinone treatment in Hgd À/À mice and patients with AKU on the abundance of urinary metabolites altered in Hgd À/À vs. Hgd þ/À mice.
The compounds that showed differences between Hgd À/L vs Hgd þ/L mice (Table 1) were examined in two additional datasets. Paired t-tests were employed to compare the abundances at baseline versus 1 week on nitisinone (4 mg/L, in drinking water) for Hgd À/L mice and 24 months on 2 mg daily nitisinone for patients with AKU. Only compounds with false discovery rate adjusted P < 0.05 pre-vs on nitisinone in mouse or human are displayed. Direction of alteration and log 2 fold change are indicated; red and blue shading indicates increased and decreased abundance on nitisinone, respectively. Where compounds were significantly different in positive and negative polarity, the result with the lowest fold change is provided. Note: no fold change indicated for hydroxymethyl-HGA in humans, as this compound was not detected for any patient on nitisinone.ND: compound not detected at baseline or on nitisinone. direct consequence of the inhibition of HPPD by nitisinone and the consequential hypertyrosinaemia. 8,9 For the first time, the present data show that targeted Hgd disruption in Hgd À/À mice induces metabolic changes upstream of HGA, despite no increase in tyrosine, hydroxyphenylpyruvic acid or hydroxyphenyllactic acid. The cause of these unexpected changes to tyrosine and peripheral neurotransmitter metabolism in untreated AKU is not clear. At the supraphysiological concentrations observed in AKU, HGA could potentially act on other enzymes of neurotransmitter metabolites and alter their activity. Alternatively, the changes could relate to an unknown feature of the disease.
In nitisinone-treated AKU, the purine metabolite changes previously reported were mainly decreased concentrations. 8 The present data indicate a more complex pattern of purine alteration in untreated AKU; increased 3,5-cyclic AMP and xanthosine, and decreased 2,3-cyclic AMP and inosine 5 0 -monophosphate. Purine catabolism is important in the homeostatic response to various states of mitochondrial oxidative stress, with shifts occurring to favour breakdown to xanthine and uric acid, the final breakdown products of purines. 44,45 The increased xanthosine and decreases in the upstream purine pathway metabolites 2,3-cyclic AMP and inosine 5 0 -monophosphate reported here for Hgd À/À mice is consistent with a shift in purine catabolism as a protection mechanism against HGA-induced oxidative stress in untreated AKU, as xanthine derivatives have antioxidant properties. 46 There is growing evidence that HGA-induced oxidative stress is an important feature of AKU (see Ref. 22 for an extensive review on this subject), and that it exacerbates ochronosis, the central AKU disease process. 47 The increased xanthosine observed here is likely a new metabolic marker of HGA-induced oxidative stress in AKU; urinary xanthosine is markedly increased in other oxidative diseases, including gout, 48 chronic kidney disease and diabetes nephropathy, 49 and it is markedly decreased following administration of nephroprotective therapy.
Decreases in TCA-related metabolites in Hgd À/À is the first indication of perturbed energy metabolism in untreated AKU. Decreases in citramalic acid, citric acid and Nacetyl-L-glutamic acid were reversed on nitisinone in Hgd À/ À , suggesting for the first time that nitisinone is at least partially restorative. Blood and urine concentrations of TCA metabolites are generally considered to directly reflect overall TCA cycle activity. 50 Inhibited TCA cycle activity could be due to overall mitochondrial biogenesis, decreased expression of genes encoding TCA cycle enzymes, dysregulation of enzyme activity or reduced substrate availability. The latter explanation seems the most likely in AKU, in which HGD deficiency prevents further metabolism of HGA to the TCA cycle intermediate fumaric acid. Decreased bioavailability of fumaric acid may then explain the decreases observed for other TCA cycle metabolites. Previous analyses have investigated potential changes to TCA cycle metabolites in AKU, but these were limited to studying the effect of nitisinone; serum concentrations of TCA metabolites succinic acid and a-ketoglutaric acid were decreased in patients with AKU on nitisinone. 9 The significance of these decreases is not known, and further studies should investigate whether the alteration to urinary TCA cycle metabolites reported here also applies to serum in untreated AKU.
Decreased urinary citrate is a well-known risk factor for kidney stone formation. It is possible that the decreased citric acid reported here for Hgd À/À mice is associated with the increased risk of kidney stones in AKU. 51 Hypocitraturia Figure 5 Summary of metabolites altered in Hgd À/À mouse urine grouped by their associated pathways. Left: the tyrosine degradation pathway showing lack of the enzyme HGD in AKU and the consequential increase in HGA. Right (boxes): observed metabolite alterations grouped by pathway; red and blue indicate increased and decreased abundance respectively. Tyrosine metabolites, including HGA and HGA biotransformation products, were elevated. Metabolites associated with the TCA cycle were decreased. A combination of increased and decreased abundance was observed for purine pathway metabolites.
is generally defined as citric acid excretion <320 mg (1.67 mmol) per day in adults. 52 The present profiling data are semi-quantitative; targeted quantitative assays are required to determine exact urinary citric acid concentrations in AKU patients and the additional therapeutic value of nitisinone in this regard.

Conclusions
In conclusion, we have shown that targeted homozygous disruption to the Hgd gene induces previously unknown metabolite changes in a complex network of pathways associated with altered tyrosine metabolism. These changes include the biotransformation of HGA by mainly phase II metabolic processes and relatively little by phase I. This observation will encourage further as yet unexplored potential treatment targets for detoxification of HGA, the culprit molecule in AKU. The wider take-home message of this study is that inherited metabolic diseases, such as AKU, present unique windows into metabolism that can uncover associations between metabolic pathways in physiology and biochemistry more generally.