Untargeted metabolomic analysis of urine samples in the diagnosis of some inherited metabolic disorders

Background. Metabolomics is becoming an important tool in clinical research and the diagnosis of human diseases. It has been used in the diagnosis of inherited metabolic disorders with pronounced biochemical abnormalities. The aim of this study was to determine if it could be applied in the diagnosis of inherited metabolic disorders (IMDs) with less clear biochemical profiles from urine samples using an untargeted metabolomic approach. Methods. A total of 14 control urine samples and 21 samples from infants with cystinuria, maple syrup urine disease, adenylosuccinate lyase deficiency and galactosemia were tested. Samples were analyzed by liquid chromatography on aminopropyl column in aqueous normal phase separation system using gradient elution of acetonitrile/ammonium acetate. Detection was performed by time-of-flight mass spectrometer fitted with electrospray ionisation in positive mode. The data were statistically processed using principal component analysis (PCA), principal component discriminant function analysis (PCA-DFA) and partial least squares (PLS) regression. Results. All patient samples were first distinguished from controls using unsupervised PCA. Discrimination of the patient samples was then unambiguously verified using supervised PCA-DFA. Known markers of the diseases in question were successfully confirmed and a potential new marker emerged from the PLS regression. Conclusion. This study showed that untargeted metabolomics can be applied in the diagnosis of mild IMDs with less clear biochemical profiles.


INTRODUCTION
Metabolomics is an emerging science which studies the complex profile of low-molecular weight metabolites present in biological samples at a specific time.It has been applied in a number of fields (environmental chemistry, plant biochemistry, microbiology and nutritional studies).It has also become an important tool in clinical research and in the diagnosis of human diseases 1,2 .The first attempt to use metabolomic tools in diagnosing inherited metabolic disorders (IMDs) was conducted by Gary Siuzdak´s group 3 .These authors applied an untargeted metabolomic approach using reverse phase capillary liquid chromatography -time-of-flight mass spectrometry based on exact mass measurements and automatic data processing.The data were processed using nonlinear alignment XCMS software 4 and METLIN online database 5 (http://metlin.scripps.edu)to find and identify metabolites differently regulated in various diseases.The concept was successfully validated in the case of two clinically profound metabolic disturbances (methylmalonic acidemia and propionic acidemia) which are characterized by a prominent biochemical profile.
The aim of this study was to determine its validity in diagnosing IMDs in urine samples using an untargeted metabolomic approach by high performance liquid chromatography (HPLC) coupled with time-of-flight mass analyzer (Q-TOF).We tested four different IMDs in which the biochemical peculiarities are not so obvious -cystinuria (CYS), maple syrup urine disease (MSUD), adenylosuccinate lyase deficiency (ADSL), and galactosemia (GALT).

Urine samples
The urine samples were taken from infants during routine diagnosic laboratory procedures.The collected urine samples were stored at -20 °C.Prior to preparation, the samples were allowed to thaw at room temperature.
Healthy control (n=14) and patient (n=21) urine samples including the 4 inherited metabolic disorders above (Table 1) were analyzed.The diagnoses had been confirmed by biochemical, enzyme or molecular-genetic analyses in all the patients (Table 2).Samples with the same number were from the same patients taken at a different time (assigned small letters).

Untargeted metabolomic analysis
Urine samples were diluted to a creatinine concentration of 0.5 mmol/L with mobile phase and 2 μL of the diluted urine were injected and analyzed by liquid chromatography coupled with mass spectrometry.For separation, a modification of a published method was used 6,7 .Separations were performed on aminopropyl column (Luna 3 μm NH2, 2 x 150 mm, Phenomenex, Torrance, CA, UHPLC Agilent 1200 Series).Gradient elution at flow rate of 0.25 mL/min was set using a program with the mobile phase A (20 mmol/L ammonium acetate, pH 9.45) and mobile phase B (acetonitrile) as follows: 0-15 min, 85% B to 15% B; 15-25 min, 15% B; 25-25.1 min, 15% B to 85% B. Between runs, the system was equilibrated with 85% B for 9 min.Total analysis time was 35 min.Detection was performed with Agilent G6520A Q-TOF fitted with electrospray ionisation in pos-itive mode.Electrospray ionisation source settings were: V Cap of 3000 V, skimmer of 65 V and fragmentor of 100 V.The nebulizer was set at 45 psi and the nitrogen drying gas was set at a flow rate of 8 L/min.Drying gas temperature was maintained at 325 °C.The data were acquired in continuum mode at a scan rate of 1.0 Hz in a mass range of m/z 70-1100 and with mass resolution of 20.000.

Data processing and statistical analysis
Data were extracted from "mzdata" raw files using the XCMS package in R software 8 (www.r-project.org).After extraction, the data were baseline corrected, normalized by the total ion count, mean centered on zero and scaled to unit standard deviation (auto-scaled).Data were statistically processed by principal component analysis (PCA) and principal components discriminant function analysis (PCA-DFA) using R software.Features most important in discrimination were found from the partial least squares (PLS) regression.Metabolites were identified using automatic workflows PUTMEDID_LCMS (ref. 9), METLIN,  *The patient samples with the same number were taken from the same patients in different time (assigned by small letter).F, female; M, male; unk, unknown; Gly, glycine; LOD, limit of detection; Leu, leucine; Ile, isoleucine.
and HMDB (http://www.hmdb.ca)databases.Identity of metabolites was confirmed by comparison with corresponding standards.

RESULTS AND DISCUSSION
The total dataset after XCMS processing comprised a total of 1492 features.First, data were statistically processed using unsupervised PCA.All the patient samples were distinguished from the controls in the PCA analysis (Fig. 1).Second, PCA-DFA, a supervised method, was applied.Distinction of all five groups (controls and patients with CYS, MSUD, ADSL and GALT) was unambiguously confirmed (Fig. 2).
A list of twenty most significant features differentiating the diseases from normal urine samples was created for each disorder based on the PLS regression.The most unambiguous results were obtained for patients with cys-tinuria.This defect is characterized by several biochemical markers (Table 1), which can be easily analyzed by the method described.The majority of these markers were confirmed (Table 3).A feature with retention time in dead volume (measured m/z of 445.2397) listed first in Table 3 was not successfully identified.A feature with m/z of 196.0790 was identified as [M+NH 4 ] + of cysteinylglycine (Cys-Gly) based on exact mass.Identification of this compound was subsequently confirmed by analyzing the appropriate standard.
One of two known markers -SAdo (Table 1) -was confirmed in patients with an ADSL deficiency.This was markedly increased in patient samples compared to control samples.Features corresponding to SAdo were among the three most discriminatory compounds.Specifically, they were identified as SAdo (molecular ion with m/z of 384.1156), its isotope (m/z of 385.1175), and its fragment (m/z of 340.1329).This was observed at the same retention time of 1050 s, consistent with the appropriate  standard.The second marker of this disease, SAICAr, was not identified in the samples.One reason for this may have been be its instability in the ion source 10 .Galactosemia is characterized by increased galactose and galactitol in urine (Table 1).In this study, we analyzed urine samples from treated adolescent and adult patients (Table 2).In treated patients, galactose and galactitol were reduced to normal or slightly above normal levels 11 .The intention of including GALT patients was to elucidate novel biomarkers of the disease.This was however not the case and patients were not distinguished by traditional markers.
Urine samples from patients with MSUD are characterized mainly by increased levels of various organic acids (Table 1).Measurements proceeded in positive mode.Hence their analyses were not sufficiently sensitive.Nevertheless, leucine and valine are also important markers of this disease.A fragment of leucine (m/z of 86.0954) was found among ten most significant features.It was substantially increased in the patient samples in comparison with controls.Its retention time fitted the appropriate standard.

CONCLUSION
This study tested the method of untargeted metabolomics in conjunction with unsupervised (PCA) and supervised (PCA-DFA) data processing for detecting patients with specific IMDs.Reported data on this method used samples from patients with diseases with an extreme biochemical profile to confirm the validity of this concept.We selected four diseases with mild biochemical abnormalities -cystinuria, maple syrup urine disease, adenylosuccinate lyase deficiency, and galactosemia.The approach was successful: patient samples were discriminated from controls by appropriate metabolites.

Table 1 .
Summary of studied defects with their urine markers.

Table 2 .
Summary of analyzed urine samples from patients with IMDs.

Table 3 .
List of ten most significant features differentiating patients with cystinuria from normal urine samples which were identified using partial least squares regression.