Defining the Healthy Infant Metabolome: Liquid Chromatography Tandem-Mass Spectrometry Analysis of Dried Blood Spot Extracts from the Johns Hopkins All Children’s PREDICT Birth Cohort Study

Abstract

Newborn screening using Dried Plasma Spots (DPS) offers pre-analytical advantages over conventional cards for plasma-associated targets of interest. Herein we present DPS-based methods for measuring metabolites using a 400+ compound LC-MS/MS library. Quality assurance reduced this library to 134, and from these, 30 compounds determined the normal newborn reference ranges.

INTRODUCTION

The dried blood spot (DBS) sampling matrix is a ubiquitous staple of state-to-state newborn health screening programs as it is widely available, cheap, and readily amenable to quality assurance programs, such as the Centers for Disease Control and Prevention’s (CDC) Newborn Screening Quality Assurance Program (NSQAP). Newborn screening is widely regarded as one of the most successful public health programs in the United States with approximately 4 million infants tested annually for inborn errors in metabolism¹. Screening programs using DBS have evolved over five decades of use, beginning with manual bacterial inhibition assays for early identification of phenylketonuria and other inborn errors of metabolism², to more sophisticated radioimmunoassay testing platforms for congenital hypothyroidism, congenital adrenal hyperplasia, cystic fibrosis, and hemoglobinopathies³. Contemporary advances using genomic sequencing technologies⁴ and tandem mass spectrometry⁵ now enable DBS-based screening for other rare diseases, where early diagnosis, often prior to the onset of symptoms, provides considerable benefits to newborns if intervention is provided as close to birth as possible.

Conventional newborn screening with DBS begins with the addition of capillary blood droplets, derived either from heel- or finger-stick within 24–48 hours after birth, onto a solid, cellulose based filter card which is then passively dried in ambient conditions. In this context, pre-analytical issues emerge as blood in a non-uniform manner dries in the presence of cells, and differential blood cells lyse and introduce contaminants⁶. Hematocrit variation⁷, as well as the inconsistency between successive blood drops from finger-sticks often used to increase total spotting volumes⁸, add to pre-analytical variation. Cells present in spotted blood may continue to undergo metabolism and respond to clotting factors⁹. These pre-analytical variables are difficult to measure and pose challenges to the utility of conventional DBS cards in ultra-sensitive testing platforms, such as therapeutic drug monitoring using mass spectrometry (MS) and other discovery applications¹⁰. A recently developed, commercially available DBS matrix (Noviplex™) separates volumetric plasma samples from blood drops at the point of DBS collection using the top layer of the card to trap cells, wherein plasma travels by capillary action to a separate disc beneath it. Plasma is thus isolated without the presence of whole or lysed cells, potentially mitigating the pre-analytical challenges described above in the use of highly-sensitive testing in DBS, herein referred to as dried plasma spots (DPS).

In a comprehensive review of the literature in 2017, we identified over two thousand analytes potentially detectable in DBS using a variety of different analytical methods¹¹. Recently, with the development of highly sensitive large multiplex liquid chromatography tandem mass spectrometry (LC-MS/MS) assays, the diversity of compounds found in whole blood spots on Whatman Filter cards, the traditional card type in most inborn errors of metabolism screening, using a single method has surpassed four hundred compounds in a recent analysis¹². Herein, we describe a workflow that combines sample collection on Noviplex Duo Cards, extraction of plasma from the lower discs, and subsequent analysis of the extracts using LC-MS/MS with a pentafluorophenylpropyl (PFPP) stationary phase with a 400 compound multiplex assay developed with authentic chemical standards. This workflow was applied to the Johns Hopkins All Children’s Prospective Research on Early Determinants of Illness and Children’s health Trajectories (“PREDICT”) birth cohort study¹³.

The objectives of the present work were to develop and describe a standardized procedure for infant DBS collection, storage, and metabolomics analysis suitable for broad implementation, and, using this approach, to derive normative values for a diverse array of analytes within the healthy infant metabolome.

METHODS

A detailed description of all methods are provided online. De-identified samples from the PREDICT birth cohort study¹³ were used in this study. Our study did not include any exclusion criteria related to premature birth as mothers were recruited early in their pregnancy. If mother’s delivered at <32 weeks, they were removed from the study. Median gestational age at birth of the infants in the study population was 39 weeks, with an observed range (36–41 weeks).

Sample collection kits were distributed to enrolled maternal participants containing a 3” by 6” aluminum canister sealed with a 3” aluminum lid with external threading and a polyvinyl chloride (PVC) coating for a hermetic seal (Elemental Container, Union, NJ). This canister held a Noviplex™ DUO Plasma Prep Card (Novilytic, West Lafayette, IN) in its original packaging, along with a 5.0-gram molecular sieve desiccant packet (Impak, Los Angeles, CA) to preserve the cards for later use. This kit was used for collection and transport of DBS specimens from the point of care at the JHACH newborn nursery to (and transient storage in) the Johns Hopkins All Children’s Pediatric Biorepository, located adjacent to the hospital.

At the time of heel stick performed for routine clinical care (i.e., state-mandated DBS newborn screening) by trained nurses in the newborn nursery nurses ≥ 24 hours after birth, two additional drops of fresh blood (estimated by the manufacturer to be approximately 60 ≥ μL) were collected for the present research during the period of 1/23/19 to 9/20/19, via application to the plasma separation DPS card in accordance with manufacturer specifications. Following a 3-minute ambient incubation, the top layer of the plasma separation DPS card was removed by the collection team, allowing the plasma to physically separate from blood cells by capillary action onto dual discs directly underneath the top layer. The plasma-separated DPS discs were subsequently placed back into the sealed transport/transient storage canister, and transported via pneumatic tube system from the newborn nursery to the Johns Hopkins All Children’s Pediatric Biorepository.

Upon receipt in the Biorepository, plasma spots were removed and separated from the DPS card as distinct discs by trained biorepository technicians using a clean wooden stick. Isolated discs were then placed inside an internally-threaded 1.4 mL polypropylene cryovial (Micronic, Lelystad, Netherlands), the cryovial was immediately purged with a 3-second nitrogen gas flush while held upright, and then sealed with the accompanying screw cap, which bears a silicon gasket. Cryovials were immediately accessioned into a −80°C semi-automated freezer (SAM, Hamilton Storage, Franklin, MA) with NIST-certified continuous electronic temperature monitoring.

For analysis, cryovials containing plasma discs were removed from long-term storage and transferred to the Johns Hopkins All Children’s Molecular Determinants Core. Each disc was extracted in 500 μl of 50/50 methanol (Optima™ LC/MS Ref# A456, Fisher Chemical, Jessup, MD)/water (Optima™ LC/MS Ref# W6, Fisher Chemical, Jessup, MD) mix and sonicated for 5 minutes before aliquoting. Each extract was split into two analytical aliquots of 200 μl each. The remaining 100 ul of extract was pooled to create two, 200 μl quality control (QC) aliquots. Aliquots were dried in a vacuum concentrator (SpeedVac Vacuum Concentrator, Savant, ThermoFisher Scientific) for interim storage at −80°C. Before analysis, one analytical and one QC aliquot for each disc were reconstituted in 50 μl of water with heavy internal standards (listed in Supplemental Table 1) for normalization. Osmolality for each sample was measured on an osmometer (OsmoPRO® Multi-Sample Micro-Osmometer, Advanced Instruments, Norwood, MA) in accordance with manufacturer instructions, using 20 μl from each aliquot.

High-performance liquid chromatography (HPLC) was performed using a Shimadzu HPLC comprised of a SIL-30ACMP 6-MTP Autosampler, and Nexera LC-30AD HPLC Pumps (Shimadzu, Columbia, MD). Mass spectrometry targeted assays employed a Shimadzu 8060 triple quadrupole mass spectrometer equipped with an electrospray Ionization source used in both positive and negative mode. Instrument settings are provided in Supplemental Table 2a. Pentafluorophenylpropyl chromatography was used to resolve small molecule metabolites, using a 150mm x 2.1 mm internal diameter column from Suppleco (Bellefonte, PA), with 3 uM particle size 3 uM (see Supplemental Tables 2b and 2c for details on gradient concentrations and durations).

Each batch was run with a System Suitability Quality Control (QC), which was created from commercially available pooled human plasma (Innovative Research Inc. Novi, MI), to validate instrument performance. Asparagine, Methionine, and Tryptophan (representative metabolites spread in time through the chromatogram) were evaluated for retention time, raw signal, and signal to noise to ensure QC suitability criteria were met. Data analysis was performed with MetaboAnalyst Software 5.0. (Edmonton, AB, Canada). The data were reduced and centered by Pareto Scaling (mean-centered and divided by the square root of the standard deviation of each variable) for multivariate analysis. Missing values were replaced by a small value (one fifth of minimum positive value in the original data, per metabolite). Regression analysis was performed by analysis of calibrators from 460,000 fmol to 8 fmol on column using a linear regression. NoviPlex DuoTM cards have an indicator of when sufficient blood volume is placed; however, upon separation of the top and bottom layer of the cards, there is no visual indicator demonstrating sufficient plasma reached the residual bottom disc. As an additional quality control check, we performed an analysis of osmolality on the specimens. Values were divided by mOsm to normalize for amount of material spotted to the DPS disc. For hypothesis testing related to specific analytes, our general approach is to use standard curves consisting of at least external authentic chemical standards for the analytes targeted in our hypothesis to obtain absolute quantitation relative to a reference standard. Targeted results were generated as femtomole (fmol) for each sample, and median and IQR values were reported for each analyte of interest specified in the tables (Tables 1 and S6). These values were converted to a μmol/L unit for more direct comparisons with published clinical reference ranges by dividing by the volume of plasma (3.8 μL) deposited per disc as a value established by the manufacturer. Data were analyzed using a non-parametric approach to construct reference values (2.5^th–97.5^th percentiles). Outliers were not excluded, unless due to previously defined analytical or biological reasons.

Table 1:

Normal research reference ranges.

Compound	Median Plasma (fmole)	IQR	Normative Range (2.5th-97.5th percentile)
Amino Acids
Alanine	839	576	155	2705
Asparagine	1724	611	991	2795
Creatine	6511	1601	3244	8004
Creatinine	3091	683	1797	4124
Glutamic acid	2101	694	1406	3337
Glutamine	3230	1092	1606	5638
Glycine	7026	2321	4429	10739
Histidine	2061	1190	869	3486
Leucine	1873	972	1102	3167
Methionine	1896	891	838	2861
Methionine Sulfoxide*	3386	1314	1579	11659
Proline	4938	1622	2436	7330
Serine	6076	3239	3771	11844
Taurine	4958	2209	2923	7335
Threonine	326	203	119	682
Tryptophan	663	280	342	1065
Tyrosine	3631	1530	1808	7413
Valine	3687	1543	1972	5917
Carboxylic acid
Orotic acid*	8467	1086	6664	10249
Dihydroorotic acid*	1573	209	1323	1960
Kynurenine*	139	22	82	256
Carnitines
Acetylcarnitine	319	99	167	510
Carnitine	536	357	274	867
Bile Acid
Glycocholic acid*	16	14	6	70
B Vitamines
Nicotinamide	1555	737	884	3701
Pantothenic acid	488	499	214	1633
Nucleic Acid
Methyladenosine*	102	28	50	119
Oligopeptide
Ophthalmic acid*	415	196	126	868
Phospholipid
Sn-Glycero-3-phosphocholine*	667	197	369	938
Stimulants
Caffeine	4684	8328	231	18783
Paraxanthine*	796	949	98	2423

^*Denotes compounds for which targeted metabolic testing is not currently available

NoviPlex Duo™ manufacturer’s instructions suggest that most analytes are stable for up to 7 days at room temperature storage. As the stability of analytes may vary depending on many factors, we assessed the impact of room temperature storage time on analytes measured by LC-MS/MS prior to freezing. Following peak selection criterion applied equally to all samples regardless of the duration of storage temperature, samples were divided into two groups in silico: Group 1: Less than seven days of room temperature storage (n=20); and, Group 2 (n=10): More than seven days of room temperature storage. Given the rather small sample size and possibility of non-parametric distribution of data, intergroup differences were tested via a Mann Whitney U test. As the goal of this study was to assess our workflows and sample stability at room temperature prior to low temperature storage, we chose 32 representative small molecules out of the 134 reported from our panel, representing ~20% of our analytical platform.

RESULTS

Thirty samples were collected from healthy normal infant patients during the study period for the purpose of establishing normative ranges of analytes determined using our multiplex LC-MS/MS assay from NoviPlex Duo™ cards. Of the 30 samples, the time from collection and storage at room temperature to the transfer of the plasma containing discs to ultra-low temperature storage ranged from less than 1 day to 32.9 days with a mean of 7.9 days with a standard deviation of 7.12 days.

Samples were extracted as described in the methods and run in a randomized single batch of 30 samples using our LC-MS/MS panel. After data processing and adjustments, we found that 134 analytes passed our initial QC criterion. For the purposes of testing the effect of duration of room temperature storage prior to freezing, we divided our samples in silico into two groups. Group one contained samples stored for less than seven days (in accordance with manufacturer’s recommendations) at room temperature prior to low temperature storage (n=20) and group two consisted of samples stored for greater than seven days at room temperature prior to low temperature storage (n=10). Analysis was performed on raw signal intensity as described above. Our results showed that 10 of the 134 analytes measured were impacted by storage time at room temperature prior to cold temperature storage (Supplemental Table 3).

Osmolality ranged from 40 mOsm/kg H₂O to 88 mOsm kg H₂O with a median value of 60 mOsm/kg H₂O, a mean value of 62.6 mOsm/kg H₂O, standard deviation of 13.3 mOsm/kg H₂O (Supplemental Table 4). Our osmolality results showed that no significant differences (mean difference of 0.81 mOsm) were associated with time (p=0.71 Mann Whitney), suggesting that any differences in osmolality were due to actual physiological differences in subjects or variability in blood volume collected.

We performed unsupervised machine learning on the remaining 20 specimens stored for less than seven days at room temperature in comparison with a) blank injections (n=16), b) sample independent quality control specimens (n=8), c) pooled specimen quality control specimens (n=8); and, d) individual specimens (n=20) to obtain a qualitative overview of differences between these groups. As expected due to a priori established criteria for variability, we obtained tight clustering of our quality control indicators (group a, b, and c). Group d exhibited spread in Euclidian space as compared to groups a and b, whereas group c was located within group d (Supplemental Figure 1.). Note that group c is a pool of all 30 specimens run with eight technical replicates, whereas group d represents the remaining 20 individual samples stored <7 days. Interestingly, a full separation of groups c and d were obtained by unsupervised analysis, indicating that the differences contributing to group separation survived a 1/3 dilution factor (10 > seven days RT stored samples out of 30 total samples) (Supplemental Figure 2). In fact, the analytes that were significantly different between group c and d were completely overlapping with the compounds listed in Supplemental Table 3 above, with the exceptions of Carnitine, Phosphoethanolamine, Isoleucine, Serine, Valine, and Taurine, indicating the effect of temperature was a profound contributor to variability.

Table 1 shows the median plasma values, inter-quartile ranges and normative values for the compounds of interest in our random sample of 20 clinically normal pediatric patients. Figure 1 demonstrates the same data, graphed as box and whisker plots, depending on scale (A-D). Supplemental Table 6 shows the remaining 102 reported compounds; instead of concentration, raw instrument intensity values are reported.

Figure 1.

Box and whisker plots demonstrating seven analytes with interquartile ranges (IQR) between 0 – 10 μmol/L (A), fourteen analytes with IQR range of 0 – 25 μmol/L (B), ten analytes with IQR range of 0 – 80 μmol/L (C), and glutathione oxidized, with an IQR range of 0 – 2000 μmol/L (D).

DISCUSSION

In this study, we identified and defined normative values among 32 analytes in the healthy newborn metabolome, via tandem LC-MS/MS analysis of capillary venous blood collected via heel-stick onto plasma-separation dried blood spot cards. Our findings are generalizable to a methodology that employs room-temperature storage less than 7 days prior to extended storage at −80C, osmolality as a surrogate normalizing measure for volumetric inconsistencies in blood spots, and highly sensitive mass spectroscopy in combination with the use of heavy internal standards. Our findings are among the first to define the healthy newborn metabolome in a pragmatic and scalable dried plasma spot (DPS) approach, at a time in which newborn screening programs look to evolve capabilities and improve laboratory quality, particularly with tandem MS^14–15. It is critical to note that these values are not validated for clinical use; however, they provide a useful reference range for research studies where cross-platform validation can be performed.

A key feature of our study includes the timing of our research sample collection with the state mandated newborn screening DBS collection (> 24 hours after birth but no later than 48 hours after birth) as metabolic profiles shift dramatically if the sample collection occurs outside of this window, as research collections often do. Furthermore, a previous study demonstrated potential differences in amino acid stabilities between cotton and cellulose filter cards when exposed to varying environmental conditions, particularly in metabolites known to be sensitive to humidity and temperature, such as histidine, tryptophan, arginine, lysine, citrulline, ornithine, asparagine, and glutamine¹⁶. We anticipate that similar stability differences exist in DPS cards and hence our collection and storage procedure (low oxygen and ultra-cold storage using barcoded cryovials) serves to mitigate variations in measured values of these and other analytes stemming from pre-analytical conditions, while also streamlining disc management practices for alignment with automated storage and analytics. This is not to say that considerable variability in some markers exists. For example, oxidized glutathione (GSSG) stability is very prone to variation associated with handling of samples (see Enns & Cowan, J Clin Med, 2017; Enomoto, et al. Biomed Chrom, 2020). Indeed, we observed a high degree of variability in GSSG indicating that there is still much room for improvement in methodologies as they relate to analytes highly susceptible to oxidation.

Notably, a recent study also used LC-MS/MS to establish reference intervals for 25 amino acids measurable in DBS cards collected from the Neonatal Cohort (aged 0 to 4 days) of the National Children’s Study (NCS) stored desiccated at −80°C ¹⁷. We observe distinctive differences between panels, particularly in overlapping amino acids wherein our reported median values are lower and the normative ranges much narrower. We hypothesize these differences suggest the NCS dried blood spot collection has a wider variation in blood spot volumes or other attributive pre-analytical variation, as the authors did not attempt to remove such outliers from their data set. We further note the detection of caffeine and related metabolite paraxanthine in our panel, as a potential surrogate marker for caffeine-based therapeutics provided to mothers after delivery and likely shared via breast-milk given the short half-life of caffeine. As there were no premature births in this study, presumably, caffeine was administered as part of an unidentified treatment or injested by mother’s before birth. Our study has several limitations. Notably, while study collections were done under a standardized protocol, it is possible that there was some variability in the amount of time before the top layer was removed from the bottom spots. We did perform studies on the amount of blood spotted in an experiment where 10 to 80 uL of blood was spotted in 10 uL increments. We found that the most reproducible range was withing the 30–60 uL range. We found significant differences in analytes above or 60 uL and below 30 uL of blood spotted (data not shown, - 3 replicates of each duo card, so 6 DPS spots total).

We observed significant variability in our analytes, likely the relatively small sample size (n=20) is in part responsible for the rather wide ranges in the normative values we report; however, physiological variation in some of the measured analytes likely also contributes. It is clear that this methodology is different than venous or DBS blood draws. In fact, our data show considerably lower levels of analytes in comparison to DBS studies for the analytes that overlap (PMID: 29854440), a finding that is consistent with DBS spots as they relate to plasma (PMCID: PMC7422991). Further research must be performed before DPS cards can be considered for clinical use.

Nevertheless, the sample size of this initial study particularly limited our power to detect differences in normative metabolomic profiles based on sex or other demographic factors¹⁸. Another potential limitation was the permissible duration of ambient storage prior to storage at −80°C; however, we sought to evaluate a pragmatic collection and storage workflow suitable for potential application in less resource-rich environments. Future larger studies will permit sensitivity analyses of the potential influence of duration of ambient storage from 0 – < 7 days; however, in preliminary analyses in the present study, we did not find this to be significantly associated with levels of measured analytes.

The focus of this study was not on newborn screening or analytes on the recommended universal screening panel (RUSP) for inborn errors of metabolism (IEMs). The results are based upon a large mass spectrometry research assay that the laboratory has developed to assess overall metabolism and provide insights into targets not generally previously studied. Unfortunately many of the targets that are indeed of importance for the RUSP IEMs are not included in this assay due to the biochemical limitations of the methods. Given the positive data associated with DPS, the translation of the DPS findings to the RUSP IEMs are an important next step in the line of investigation. Notably, our panel does involve important markers that could potentially be used to detect synthesis defects in serine and creatinine biosynthesis. These pathways will be the focus of next steps in our research as the analytes are compatable with the biochemical methods used in the study.

Despite these limitations, this study establishes feasibility of a DPS collection and storage methodology, in combination with LC-MS/MS, for determination of a broad array of analytes within the healthy newborn metabolome. We report normative values for 32 metabolites with suitable stability for targeted comparative studies in larger cohort studies, wherein diagnostic and prognostic utility may be shown valuable for identifying metabolic disorders and molecular early determinants of child health and disease.