Developing a SWATH capillary LC-MS/MS method for simultaneous therapeutic drug monitoring and untargeted metabolomics analysis of neonatal plasma

Abstract

Most medications prescribed to neonatal patients are off-label uses. The pharmacokinetics and pharmacodynamics of drugs differ significantly between neonates and adults. Therefore, personalized pharmacotherapy guided by therapeutic drug monitoring (TDM) and drug response biomarkers are particularly beneficial to neonatal patients. Herein, we developed a capillary LC-MS/MS metabolomics method using a SWATH-based data-independent acquisition strategy for simultaneous targeted and untargeted metabolomics analysis of neonatal plasma samples. We applied the method to determine the global plasma metabolomics profiles and quantify the plasma concentrations of five drugs commonly used in neonatal intensive care units, including ampicillin, caffeine, fluconazole, vancomycin, and midazolam and its active metabolite α-hydroxymidazolam, in neonatal patients. The method was successfully validated and found to be suitable for the TDM of the drugs of interest. Moreover, the global metabolomics analysis revealed plasma metabolite features that could differentiate preterm and full-term neonates. This study demonstrated that the SWATH-based capillary LC-MS/MS metabolomics approach could be a powerful tool for simultaneous TDM and the discovery of neonatal plasma metabolite biomarkers.

Introduction

As an analogy to the well-known quote, “Children are not little adults”, infants are not just small children. Human organs and biochemical processes undergo substantial maturation after birth, which can greatly affect the pharmacokinetics (PK) and pharmacodynamics (PD) of drugs in neonates [1]. About 65% of drugs used in Neonatal Intensive Care Unit (NICU) are off-label uses and require dosage adjustment for neonatal patients [1]. For example, vancomycin is an off-label antibiotic widely used in NICU to treat late-onset sepsis caused by methicillin-resistant Staphylococcus aureus and coagulase-negative staphylococci [2]. The interindividual variability in vancomycin PK is high in neonates. De Hoog and colleagues reported that the half-life of vancomycin varied between 3.5 h and 10.0 h, and the clearance ranged from 0.63 to 1.40 ml/kg/min among neonatal patients [3]. Because of its narrow therapeutic window and significant nephrotoxicity and ototoxicity, therapeutic drug monitoring (TDM)-based dose individualization is essential to ensure the efficacy and safety of vancomycin treatment [4]. TDM could also be invaluable for many other medications, given that the PK profiles of most drugs have not been well characterized in neonates. Therefore, a reliable analytical method capable of simultaneously monitoring commonly used drugs in neonates is highly desirable. Besides TDM, determining the dynamic changes of endogenous small molecules at the system level (i.e., metabolomics) could lead to the discovery of metabolite biomarkers to further improve pharmacotherapy outcomes through individualized drug regimens. Thus, developing an analytical platform with the capability for both targeted TDM and global metabolomic profiling is of importance to clinical practice and research of precision pharmacotherapy in neonatal patients.

Liquid chromatography–mass spectrometry (LC-MS) has been widely used for both targeted and untargeted analyses of small and large molecules in various research settings [5–7]. Sequential Windowed Acquisition of All Theoretical Fragment Ion Mass Spectra (SWATH) technology is an emerging data-independent acquisition (DIA) strategy that isolates precursors into predefined small mass windows for fragmentation and collects all generated MS2 spectra for identification and quantification. SWATH was originally developed for global proteomics analysis and has demonstrated several advantages over conventional data-dependent and targeted data acquisition methods [8]. Unlike conventional data acquisition methods, SWATH generates a digital archive of comprehensive MS/MS data for both qualitative and quantitative analysis and allows for data re-interrogation without the need for sample re-reanalysis. Besides its wide application in proteomics, SWATH has been increasingly used in metabolomic and lipidomic research [9–11]. For example, Xiong et al. successfully applied a SWATH-based serum metabolomics method to identify pancreatic cancer biomarkers [12]. Drotleff et al. used a SWATH method for an untargeted large-scale lipidomic analysis of mouse plasma [13]. Due to the complexity of SWATH data, several sophisticated software packages have been developed for the efficient processing of SWATH data [9, 14, 15]. However, the SWATH technology has not been utilized for simultaneous TDM and untargeted plasma metabolomics profiling.

The purpose of this study was to develop and validate a SWATH-based analytical platform to enable the simultaneous analysis of global neonatal plasma metabolome and plasma concentrations of several commonly used medications in NICU, including ampicillin, caffeine, fluconazole, vancomycin, midazolam, and its metabolite α-hydroxymidazolam. The method was validated and successfully applied to the TDM and global metabolomics analysis of plasma samples from neonatal patients treated in a NICU.

Materials and Methods

Chemicals and materials

Ampicillin, caffeine citrate, fluconazole, midazolam, α-hydroxymidazolam, vancomycin hydrochloride, and the internal standards (IS) caffeine-¹³C3 and midazolam-D4 maleate, LC–MS grade acetonitrile, water, and formic acid were purchased from Sigma Aldrich (St. Louis, MO, USA). All other chemicals and agents were of the highest analytical grade commercially available.

Blank human plasma was purchased from Innovative Research (Novi, Michigan, USA). Blood samples were obtained from three preterm and three full-term neonates treated in the NICU of the University of Michigan Hospital. The study was approved by the Institutional Review Board at the University of Michigan. Plasma was separated from the whole blood after centrifugation and stored at −80°C until use.

Calibrators and quality control (QC) samples

Stock solutions of each analyte were prepared in methanol at a concentration of 1 mg/ml. Working standard solutions (100 μg/ml, 10 μg/ml, 1 μg/ml, 100 ng/ml, 10 ng/ml) were obtained by diluting stock solution in methanol. Calibration standards and quality controls (QCs) were prepared by spiking pooled blank human plasma with the working solutions (Table 1). Three concentrations of QC samples were summarized in Electronic Supplementary Material (ESM) ESM1 Table S1. The FDA Bioanalytical Method Validation Guidance and the ranges of the typical plasma concentrations of the candidate drugs (Table 2) were considered when choosing the concentrations of calibrators and QCs. Metabolomics QC samples were prepared by mixing pooled blank plasma and clinical plasma samples at a ratio of 1:1.

Table 1.

Ionization parameters and retention times of the analytes and the internal standards

Compound	Retention Time(min)	Precursor ion(m/z)	Polarity	Molecular	Product ions (m/z)	Collision Energy(V)
Ampicillin	15.7	350.1	Positive	[M+H]+	192.0, 160.0, 333.1	22.0
Caffeine	15.4	195.1	Positive	[M+H]+	195.1, 138.1, 110.1	14.1
Caffeine-3C13	15.4	198.1	Positive	[M+H]+	140.1, 112.1, 141.1	14.1
Fluconazole	17.2	307.1	Positive	[M+H]+	307.1, 238.1, 220.1	20.3
Midazolam	19.2	326.1	Positive	[M+H]+	291.1, 290.1, 244.0	21.1
α-hydroxymidazolam	19.1	342.1	Positive	[M+H]+	342.1, 324.1, 203.0	22.0
Midazolam-D4	19.2	330.1	Positive	[M+H]+	291.1, 234.1, 295.1	21.1
Vancomycin	14.9	724.7	Positive	[M+2H]2+	144.1, 100.1, 724.7	41.3

Table 2.

Performance of calibration curves, LLOQ, LLOD, and typical plasma concentration ranges of the targeted compounds

Compound	Calibration Curve			LLOQ (ng/ml)	LLOD (ng/ml)	Typical concentration range (μg/ml)
	Slope	R2	Linear range
Ampicillin	0.0103	0.992	0.1–20 μg/ml	100	5	0.85–46.4 [31]
Caffeine	0.0039	0.999	0.5–20 μg/ml	500	0.5	5–20 [32]
Fluconazole	0.0254	0.997	0.2–20 μg/ml	200	1	0.5–14 [33]
Midazolam	0.0291	0.999	0.02–50 μg/ml	20	1	0.08–3.2 [34]
α-hydroxymidazolam	0.0244	0.994	5–500 ng/ml	5	1	0.008–0.062 [35, 36]
Vancomycin	0.0028	0.998	0.002–50 μg/ml	2	0.5	0.5 −60 [37]

Sample preparation

For calibrators, 25 μl of pooled blank human plasma was mixed with 25 μl of working solutions, followed by the addition of 150 μl of methanol containing 100 ng/ml IS. For clinical samples, 25 μl of plasma was mixed with 175 μl methanol containing the same amount of IS. Caffeine-¹³C3 was used as the IS for the quantification of ampicillin, caffeine, fluconazole, and vancomycin, and midazolam-D4 maleate was the IS for midazolam and α-hydroxymidazolam quantification. Samples were vortexed thoroughly for 10 min and centrifuged at 20,000 rcf for 10 min at 4°C to remove the precipitated proteins. The resulting supernatant was transferred to a new Eppendorf tube and was vacuum dried in a SpeedVac SPD1010 concentrator (Thermo Scientific, Hudson, NH). Samples were then reconstituted in 50 μl of water/acetonitrile mixture (4:1, v/v) and vortexed for 10 min. After centrifugation at 20,000 rcf for 10 min at 4°C, 0.5 μl of the supernatant was injected into an LC-MS/MS system for analysis.

SWATH data acquisition

A SWATH-based LC-MS/MS method was established for both targeted quantification and non-targeted metabolomics analysis. The LC-MS/MS system consisted of a TripleTOF 5600 plus mass spectrometer (Sciex, Framingham, MA) coupled with a Digital PicoView 450 nanospray ion source (New Objective, Woburn, MA) and an Eksigent 2D plus LC system (Eksigent Technologies, Dublin, CA). An ACQUITY UPLC M-Class Peptide BEH C18 column (130 Å, 150 μm × 100 mm,1.7 μm, Waters, Milford, MA) was used for the chromatography separation at 40°C. The mobile phase A and B were water and acetonitrile, respectively, and both mobile phases contained 0.1% (v/v) formic acid. The mobile phase was delivered at a flow rate of 1 μl/min with the following gradient: the gradient started with 3% of B and was kept for 10 min, then linearly increased to 100% at 40 min and held for 3 min, and then changed back to 3% in 1 min and held till the end of the 50 min gradient.

The SWATH acquisition included an MS1 full scan at an m/z range of 100–1,250 Da and 60 variable precursor isolation windows calculated by the Sciex SWATH Variable Window Calculator (ESM2). For each SWATH isolation window, the MS1 and MS2 accumulation times were 150 ms and 30 ms, respectively, resulting in a cycle time of 2 seconds. The average peak width of analytes was about 30 seconds, and thus, approximately 15 data points were collected for each peak. Data were acquired under a positive electrospray ion (ESI) mode. Collision energy (CE) voltage was automatically optimized by the acquisition software (Analyst TF 1.7) for each SWATH window with the CE spread (CES) set at 15 V. The declustering potential was 100 V, the voltage of the spray was set at 3,800 V, and the temperature of the interface heater was 200°C. For the gas settings, the ion source Gas 1 was 52, the ion source Gas 2 was 0, and the curtain gas was 30.

Method validation

linearity and lower limit of quantification (LLOQ)

The linearity of analyte response was evaluated based on the correlations between the analyte to IS peak area ratios and the nominal analyte concentrations of the calibrators. A weighting factor of 1/x was used for calculation. The LLOQ was defined as the lowest concentration with a signal-to-noise ratio greater than 10, a coefficient of variation (CV%) ≤ 20%, and accuracy between 80% and 120% of the nominal concentration. A calibration curve was accepted when the back-calculated concentrations of calibrators were within 100 ± 15% of the nominal values (except for 100 ± 20 % for the LLOQ).

Intra-batch and Inter-batch accuracy and precision

Intra-batch accuracy and precision were assessed by analyzing five replicates of QC samples at three concentrations (QC-Low, QC-Medium, QC-High). Accuracy was evaluated based on the differences between the measured concentrations and the nominal concentrations. Precision was determined by calculating the CV% of the five replicates. Inter-batch accuracy and precision were assessed by analyzing the QCs from three independent batches. The acceptance criteria for the Intra-batch and Inter-batch accuracy and precision were that the measured concentrations should be within 100 ± 15% of the nominal concentrations, and the CV% should be no more than 15%.

Stability, matrix effect, and extraction recovery

Autosampler stability, freeze-thaw stability, matrix effect, and extraction recovery were assessed for each analyte using QC samples at three concentrations. Autosampler stability was studied after the extracted analytes were placed in the autosampler (4°C) for 24 hours and 48 hours before injection. For freeze and thaw stability, samples were analyzed after three consecutive freeze-thaw cycles, during which the samples were frozen at −80°C and thawed at room temperature. Matrix effect was evaluated by comparing the peak area of the analytes dissolved in the mobile phase to that of the same concentrations of analytes spiked in the post-extracted plasma. Extraction recovery was determined by comparing the peak areas of the analytes exacted from blank plasma samples spiked with the analytes (i.e., pre-spike) with the peak areas of the analytes spiked in blank plasma extraction eluent (i.e., post-spike).

Data processing and statistical analysis

The workflow of the SWATH data acquisition and analysis is summarized in Fig 1. The Skyline software (version 20.1, University of Washington, Seattle, WA) [16] was used for the targeted analysis. The sum of MS1 and MS2 peak areas were utilized for quantification, and the analyte concentrations were calculated according to the analyte/IS peak area ratios. For untargeted metabolomics analysis, the MS-DIAL software (version 4.16) was used for peak detection, alignment, MS2 information extraction, relative quantification, and metabolite annotation [17]. The parameters of MS-DIAL are listed in ESM1 Table S2, blank matrix samples were used for background subtraction in the MS-DIAL. Database-based metabolite annotations were conducted with a public metabolomics library (http://prime.psc.riken.jp/compms/msdial/main.html#MSP) containing 13,303 unique compounds. Following the MS-DIAL analysis, the peak areas of all the metabolite features in each sample were exported. Metabolite features which were detected in over 80% of the samples and showed a relative standard deviation (RSD) below 30% in the QC samples were included in data analysis. The data were then normalized to the average level and transposed before being exported to the SIMCA-P software (version 12.0, Umetrics) for principal component analysis (PCA) and partial least-squares discriminant analysis (PLS-DA).

Fig.1.

Workflow of the SWATH capillary LC-MS/MS method for simultaneous therapeutic drug monitoring and untargeted metabolomics analysis

Results

Method validation

As presented in Fig. 2, the chromatographic performance was acceptable for all analytes in neonate plasma samples under the present LC conditions. The chromatograms of blank plasma containing the analytes at the concentrations of LLOQ and of blank matrix demonstrated excellent sensitivity and selectivity of the assay (ESM Figure S1, S2). The ionization parameters and retention times of the targeted analytes and the IS were listed in Table 1. The retention times of the analytes were between 14 min and 20 min, and the collision energy ranged from 14.1 V to 41.3 V. The targeted method was validated for calibration curves, linearity, intra-batch and inter-batch accuracy and precision, matrix effect, extraction recovery, and stability. All the calibration curve R² values were greater than 0.99, and the back-calculated concentrations of the calibrators were within the range of 85%−115% of the nominal concentrations. The slope, intercept, R², and linear ranges of each calibration curve and the LLOQ and LLOD of the analytes were shown in Table 2. The intra-batch and inter-batch accuracy validation study showed that the measured concentrations of the QC samples ranged from 92.8% to 103.8% of the nominal concentrations (Table 3 and Table 4). The intra-batch and the inter-batch precision study demonstrated that the CVs of the QC samples were within 9.4% for all analytes (Table 3 and Table 4). The matrix effect study did not show significant ion suppression or ion enhancement for all analytes (ESM1 Table S3). The extraction recovery rates were similar among the analytes, ranging from 89.9% to 113.9% (ESM1 Table S3). The results also demonstrated that all analytes were stable after being kept in the autosampler for 48 hours or after three freeze-thaw cycles (ESM1 Table S3, S4).

Fig.2.

MS1 and selected MS2 chromatograms of the targeted analytes obtained from neonate samples.

Table 3.

Intra-batch precision and accuracy

	QC low		QC medium		QC high
	Precision (%)	Accuracy (%)	Precision (%)	Accuracy (%)	Precision (%)	Accuracy (%)
Ampicillin	7.0	103.3	5.3	93.6	9.4	101.7
Caffeine	5.4	102.6	8.0	99.6	1.7	102.4
Fluconazole	4.9	98.9	5.0	99.4	8.3	99.0
Midazolam	7.6	102.4	11.5	101.6	8.9	92.8
α-hydroxymidazolam	4.7	101.8	5.0	98.8	3.7	105.7
Vancomycin	5.2	100.3	9.4	96.6	2.5	103.8

Table 4.

Inter-batch precision and accuracy

	QC low		QC medium		QC high
	Precision (%)	Accuracy (%)	Precision (%)	Accuracy (%)	Precision (%)	Accuracy (%)
Ampicillin	6.4	102.6	7.5	97.5	8.7	103.1
Caffeine	4.4	99.2	7.9	100.8	7.8	99.7
Fluconazole	8.1	103.4	5.8	99.8	6.5	99.7
Midazolam	6.4	100.1	8.7	102.2	9.1	97.2
α-hydroxymidazolam	5.9	102.6	5.9	97.2	6.8	102.2
Vancomycin	7.3	98.4	8.1	100.5	8.1	101.5

Untargeted metabolomics analysis

For the untargeted metabolomics analysis, a total of 2,245 metabolite features were detected from the human plasma samples. Among these metabolite features, 1,055 were annotated by the MS-DIAL software according to their accurate mass and MS/MS spectra. The distribution of the MS¹ ion intensity of the annotated metabolites was presented in ESM1 Fig. S3, showing that most of the annotated metabolites had ion intensity greater than 1 × 10⁴ cps. The annotated plasma metabolites were classified into various classes of metabolites (Fig. 3A), and the distribution of the annotated metabolites is in agreement with the previous reports [18, 19]. The MS¹ and MS² peaks of the representative annotated metabolites phenylalanine, and LPC 18:0 are shown in Fig. 3B and Fig. 3C, respectively. The measured MS2 spectra were matched with the reference metabolite spectra, which increased the credibility of the annotated metabolites. The reproducibility of this untargeted metabolomics method was evaluated by measuring four metabolomics QC samples. Fig. 3D showed the RSD distribution of all the detected features in the four QC samples, and most of them are within 30%. The PCA analysis showed that the four QC samples were clustered closely (within 2 SD, ESM1 Fig. S4), indicating that the method was highly reproducible.

Fig.3.

Profiles of the untargeted metabolomic analysis method. (A) Classification and relative abundance of the annotated metabolites (B) Example of identified metabolites phenylalanine, left panel: overlapped MS1 and MS2 peaks; right panel: deconvoluted MS/MS spectrum from the plasma samples (blue) matched against database spectrum (red). (C) Example of identified metabolites LPC 18:0, left panel: overlapped MS1 and MS2 peaks; right panel: deconvoluted MS/MS spectrum from the plasma samples (blue) matched against database spectrum (red). (D) Distribution of RSD (%) of all the features in QC samples. The percentage of compound numbers within the corresponding %RSD range is represented by each column.

Application to clinical samples

The established SWATH metabolomics method was applied to the analysis of plasma samples collected from three preterm and three full-term neonatal patients who were treated with at least one of the five targeted drugs in the NICU of the University of Michigan hospital. As shown in Table 5, all analytes of interest were successfully measured by the assay. It is noted that the drugs detected in the study were consistent with the drug administration information in the patients’ electronic medical records, suggesting the excellent sensitivity and specificity of the method.

Table 5.

Concentrations of targeted drugs (ng/ml) measured in neonate plasma samples

Patient number	Ampicillin	Caffeine	Fluconazole	Midazolam	α-hydroxymidazolam	Vancomycin
1	18,100	ND	ND	931	131	10
2	ND	ND	ND	617	53	ND
3	4,670	ND	ND	472	46	17
4	ND	16,100	12,400	2,560	130	8,910
5	ND	4,590	7,520	212	26	8,100
6	ND	ND	ND	ND	ND	13

ND: not detected

The SWATH method was evaluated for its capability in assessing the metabolome differences between preterm and full-term neonates. The drugs measured by the targeted metabolomic analysis were removed from the metabolome dataset to eliminate the contributions of the administered medications to their metabolomic profiles. The preterm and full-term neonates were completely separated in the PLS-DA score plot (model parameters: R²X = 0.76, R²Y = 0.987, Q² = 0.324) (Fig. 4A). Due to the limited sample size (n = 3), the Q² value of the permutation plot and the CV-ANOVA test of the PLS-DA model could not be validated. In the S-plot of the PLS-DA scores (Fig. 4B), variables with absolute covariance (X axis) > 0.05 and absolute correlation (Y axis) > 0.3 were in the shaded area and were considered to be the major contributors to the classification. The loading plot of the PLS-DA scores is shown in ESM1 Fig.S5. The corresponding VIP values were calculated, and a total of 175 metabolite features were found to have a VIP value > 1. For the univariate screening of statistically different metabolite features between the two groups, the Student’s t-test with Bonferroni correction was utilized, and 106 features with P < 0.05 were retained. The fold changes of the normalized metabolite feature intensities were also calculated, and 622 features differed by greater than 20% between the two groups. The Venn diagram (Fig. 4C) shows the overlapped features reported by the three data analysis approaches, and a total of 61 features were identified by all three methods to be the metabolites differentiating between the preterm and full-term groups. Nineteen out of the 61 metabolite features were annotated using the plasma metabolomic libraries from MS-DIAL and the Human Metabolome Database (HMDB). The relative abundances of these annotated metabolites in each sample were shown in the heatmap (Fig. 4D) and violin plots (Fig. S6).

Fig.4.

Metabolomics profiles of neonate samples. (A) PLS-DA score plot of the full-term and preterm neonates. Red dots represent the three full-term neonates, and blue triangles represent the three preterm neonates. (B) S-plot plot of the PLS-DA model. The variables that contributed most to the classification were in shaded areas. (C) Venn diagram of VIP (VIP>1, n=175), P-value (P < 0.05, n = 106), and fold change (FC < 0.8 or FC > 1.2, n = 622). (D) Heatmap of 19 potential differential metabolites between preterm and termed groups.

Discussion

In this study, a SWATH capillary LC-MS/MS metabolomics method was developed for both targeted quantification and global metabolite profiling. The method was validated for linearity, sensitivity, precision, and accuracy for the targeted analytes. We demonstrated its suitability for the quantification of several commonly used medications in neonates, including ampicillin, caffeine, fluconazole, vancomycin, midazolam, and its metabolite α-hydroxymidazolam. The assay was also successfully applied to the untargeted metabolomics analysis of six plasma samples collected from neonatal patients. The preterm and full-term neonates could be readily differentiated according to their metabolite features.

Nano-flow (e.g., 300 nl/min) is the mainstay for proteomics analysis [20], whereas analytical flow (e.g., 0.2 – 1 ml/min) is most common in metabolomics research [21]. Nano-LC offers greater sensitivity but is generally less robust compared to an analytical flow configuration. In the present study, a capillary-LC system with a 150 μm ID column and a flow rate of 1 μl/ml was employed to achieve a balanced performance of sensitivity and robustness. The results were found to be highly reproducible, as demonstrated by both the validation study of the targeted analytes and the results from the untargeted metabolomics study of the QC samples.

Different from other data acquisition modes (e.g., data-dependent acquisition and multiple reaction monitoring (MRM)), SWATH generates both MS1 and MS2 chromatograms of all analytes. Therefore, SWATH quantification can be based on the peak areas of MS1, MS2, or the sum of MS1 and MS2. We compared the performances of the three quantification approaches and found that the method utilizing the sum of MS1 and MS2 signals generally outperformed the methods using either MS1 or MS2 regarding sensitivity and reproducibility. For example, when only MS1 was used for vancomycin quantification, the LLOQ was 10 ng/ml, whereas the LLOQ was 2 ng/ml when the quantification was based on the sum of MS1 and MS2. The variations of ampicillin and vancomycin calibration curves also increased when only MS1 signals were used (data not shown). In addition, when only MS2 signals were used, the R² values of the midazolam and ampicillin calibration curves decreased to 0.946 and 0.984, respectively. Accordingly, the sum of MS1 and MS2 peaks was utilized to quantify the targeted compounds throughout the study.

Several LC-MS/MS-based methods were established to quantify ampicillin [22], caffeine [23], fluconazole, midazolam, α-hydroxymidazolam [24], and vancomycin [25] in neonatal plasma. All these methods were based on the same MRM data acquisition strategy using low-resolution triple quadrupole mass spectrometers. As a comparison, this reported SWATH method utilized an emerging data-independent data acquisition technology on a high-resolution Q-TOF instrument. The generated SWATH data is a complete digital archive of all detected analytes. Thus, unlike those targeted methods (e.g., MRM), the SWATH approach allows post-acquisition data extraction. For example, if a new interest arises for a compound that was not targeted in the original study, investigators can re-interrogate the previously collected SWATH data and extract the quantitative information of the compound of interest. The flexibility of this approach was also demonstrated by two recent studies: Drotleff and collegues quantified steroid hormones in human plasma using SWATH-acquisition and untargeted profiling [26]. Sanwald et al. used both targeted and SWATH assays to quantify amino acids in human corneal epithelial cells treated with ionic liquids [27]. Moreover, the SWATH analysis generated thousands of metabolite features that successfully differentiated the preterm and full-term neonates enrolled in the study, indicating that the SWATH-based metabolomics could be a powerful tool for clinical biomarker discovery. In this preliminary study, 61 metabolite features were found to differ between the preterm and termed neonates, among which 19 were annotated. The classes of the 19 annotated metabolites include organic acids, nucleotides, nucleosides, amino acids and indole derivatives. Some of them are critical in various signaling pathways. For example, both dopamine and its precursor L-dopa play important roles in the brain activity mediated by dopamine pathways, and the plasma concentrations of the two neurotransmitters were different between the preterm and termed neonates, which is consistent with a previous report showing that the activation of the dopamine pathways in preterm patients was different from that in termed patients [28]. Apart from the signaling pathways, some identified metabolites such as creatinine are indicators of the maturation of organ functions. Indeed, creatinine plasma levels were reported to be significantly different between preterm and termed neonates [29, 30]. However, it should be noted that only six subjects were included in this proof-of-concept study, and a larger sample size is warranted in order to perform a more in-depth global metabolomics analysis. Also, chemical standards are needed for authenticating the identified metabolite biomarkers.

In sum, we developed a SWATH capillary-LC-MS/MS analytical platform capable of simultaneously analyzing the compounds of interest and the whole plasma metabolomes of neonatal patients. This method is suitable for clinical TDM for multiple medications while concurrently generating comprehensive untargeted metabolomics data for biomarker discovery and targeted post-acquisition data extraction.

Supplementary Material

1

ESM1. Supplementary experimental results

2

ESM2. SWATH isolation windows

• Simultaneously analyze targeted and untargeted metabolites in human plasma

• Preterm and full-term neonatal patients could be differentiated by the method

• A powerful tool for therapeutic drug monitoring and biomarker discovery