Untargeted and Targeted Metabolomic Profiling of Preterm Newborns with EarlyOnset Sepsis: A Case-Control Study

Sepsis is a major concern in neonatology, but there are no reliable biomarkers for its early diagnosis. The aim of the study was to compare the metabolic profiles of plasma and urine samples collected at birth from preterm neonates with and without earlyonset sepsis (EOS) to identify metabolic perturbations that might orient the search for new early biomarkers. All preterm newborns admitted to the neonatal intensive care unit were eligible for this proof-of-concept, prospective case-control study. Infants were enrolled as “cases” if they developed EOS, and as “controls”if they did not. Plasma samples collected at birth and urine samples collected within 24 h of birth underwent untargeted and targeted metabolomic analysis using mass spectrometry coupled with ultra-performance liquid chromatography. Univariate and multivariate statistical analyses were applied. Of 123 eligible newborns, 15 developed EOS. These 15 newborns matched controls for gestational age and weight. Metabolomic analysis revealed evident clustering of the cases versus controls, with the glutathione and tryptophan metabolic pathways markedly disrupted in the former. In conclusion, neonates with EOS had a metabolic profile at birth that clearly distinguished them from those without sepsis, and metabolites of glutathione and tryptophan pathways are promising as new biomarkers of neonatal sepsis.

The mass range scan was of 20 to 1200 amu, both in MS scan mode and in MSe mode to obtain the fragmentation spectrum of the variables that fall within the parameters set in the scan method in MSe. The Quality control (QC) samples and standard solution samples were used to test for reproducibility and accuracy during the analysis and were injected at regular intervals throughout the sequence, together with blank samples. To further reduce analytical variability, in accordance with an in-house protocol, samples distribution in the plate and the sequence of sample injection in the UPLC-MS were randomized, and 5/6 of the fluid resulting from the addition of eluents to the sample was excluded from the ionization process (splitting). Splitting samples prevent the risk of smudge the internal surfaces of the spectrometer itself, thus reducing its sensitivity.

S1.2 Data processing and pre-treatment
UPLC-MS data were processed by the ProgenesisQI software (Waters Corporation, Milford, U.S.A.) and two data sets were generated, one for the positive-ionization mode (POS data set) and the other for the negative-ionization mode (NEG data set). The parameters used for data extraction were optimized through the preliminary analysis processing of the QC samples. We used a filter strength of 0.25 for import raw data and a QC in the middle of the sequence as a reference for the automatic alignment of all runs in the sequence. For the peak picking the sensitivity of the automatic algorithm was set at 3, where retention time limits were between 0.4 and 8 min. The so-called Rt_mass variables (where Rt is the retention time and mass is the mass to charge ratio m/z of the chemical compound) were generated.
Variables with more than 20% of missing data were eliminated to avoid spurious statistical models generated by unrealistic combinations of the measured variables. For each variable passing such a filter, missing data were imputed with a random number between zero and the minimum value measured for the variable.
Variables with a coefficient of variation greater than 20% for QC samples have been excluded. Variables detected in the blank samples have been subtracted to the samples. The ion intensities for each peak detected were normalized, based on the calibration models obtained for the QCs with different dilution factors (1:3, 1:5, 1:7). Then probabilistic quotient normalization was applied to take into account dilution effects.

S1.3 Variables annotation
The relevant variables selected by multivariate data analysis were merged with those obtained from univariate data analysis and were annotated by searching our in-house database of commercial standards, the METLIN metabolite database, and the Human Metabolome Database to obtain a unique identification code Level 1 was assigned for the compounds with a difference of 10 ppm for m/z, 0.2 min for rt, and, where available, with collision cross-section ≥2%, with respect to the standards of our in-house database, that were performed under identical analytical conditions of the current analysis.
Instead, level 2 and 3 was for metabolites with m/z ≤10ppm respect to the online databases, and the fragmentation score ≥ 30 or < 30, respectively.

S1.4 Outlier detection
PCA has been applied to detect the presence of outliers in the untargeted metabolomic data obtained from urine samples. Specifically, T2 test and Q test have been applied with a significance level =0.05. Data have been mean-centered and 2 principal components have been considered. In figures S1 and S2 the score scatter plots and the T2/Q plots obtained respectively for the NEG and the POS data sets are reported.

S2.1 Chemicals and reagents
The chemical standards and labeled standards were purchased from: Sigma-Aldrich Corporation (Milan,  Table S1 and Table   S2. The purity of all analytes and labeled internal standards was ≥98%. Water was purified with a Milli-Q Elix purification system (Millipore, Bedford, MA, USA). High-purity MS-grade solvents (formic acid, methanol, and acetonitrile) were obtained from Fluka (Milan, Italy) and used without further purification.

S2.2 Preparation of standard solutions and calibration curves
Individual stock solutions in water or methanol with different percentage of formic acid depending on the different solubility of the compounds were used. A series of solution mixtures of desired concentrations were prepared by suitable dilutions of the stock solutions in 0.1% formic acid in water. All the stocks were stored at -20 °C.
Stock solutions of labeled metabolites were prepared as the unlabeled and diluted as required, with water 0.1% FA, to obtain a concentration of 0.05-0.1 µM for neurotransmitters, and polyamine, and 1-10 µM for amino acids and kynurenine metabolites, and used as internal standard (IS).
Calibration curves of the analytes were prepared by spiking pooled plasma, obtained from volunteers, with the diluted mixed standard solutions and IS, to the concentration ranging from 0.3 to 100 nmol/L for neurotransmitters, from 30 to 3000 nmol/L for polyamines, and from 0.05 to 250 µmol/L for amino acids and kynurenine metabolites.

S2.3 Quality controls (QC)
Two different concentrations of QC's plasma were used for precision and accuracy. Where available we used QC from chemical companies, with 2 different level concentration (Amino Acid Quality Control set, low and High, Kairos TM , Waters Corporation, Milford, MS, USA), otherwise we prepared QC by spiking pooled plasma with 2 different concentration of the analytes.
The QC's plasma were extracted 2 times and analyzed 5 times within the same chromatographic run (n=10, intraday repeatability) and for 3 distinct days (n=30, between days reproducibility) to precision and reproducibility of the analytical method, expressed as coefficient of variation (CV%).

Difference between measured and expected values of QC's plasma samples (Bias%) was used to estimate the accuracy of the analysis
The analytes with CV% and Bias% ≤ 20% were considered for targeted analysis.
Plasma calibrations curve at 5 concentrations were built for assessing linearity, expressed as R 2 .
Sensitivity, expressed as limit of quantification (LOQ, S/N≥10), was extrapolated by lowest point of calibration solution. The R 2 and LOQ for each compound were reported in Table S1 S2

S2.5 UPLC-MS analysis
The analysis was conducted using a Xevo TQ-S triple-quadrupole mass spectrometer coupled to an Acquity Specific mobile phases and injection volumes were used for the different classes of metabolites as summarized below:

Amino acids
Mobile phases consisted of waters 0.1% formic acid for phase A and acetonitrile 0.1% formic acid for phase B. Injection volume 2 µL.

Neurotransmitters associated to tyrosine and tryptophan metabolism
Mobile phases consisted of waters 0.1% formic acid for phase A and acetonitrile:methanol 90:10 0.1% formic acid for phase B. Injection volume 20 µL.
Instrument control, data acquisition and analysis were managed with MassLynx software (version 4.1, Waters). Quantification was done using the TargetLynx function of the same software.