Biofluid metabolomics of mice exposed to external low-dose rate radiation in a novel irradiation system, the VAriable Dose-rate External ¹³⁷Cs irradiatoR (VADER)

Abstract

An important component of ionizing radiation (IR) exposure after a radiological incident may include low-dose rate (LDR) exposures either externally or internally, such as from ¹³⁷Cs deposition. In this study, a novel irradiation system, VAriable Dose-rate External ¹³⁷Cs irradiatoR (VADER), was used to expose male and female mice to a variable LDR irradiation over a 30-day time span to simulate fall-out type exposures in addition to biofluid collection from a reference dose rate (0.8 Gy/min). Radiation markers were identified by untargeted metabolomics and Random Forests. Mice exposed to LDR exposures were successfully identified from control groups based on their urine and serum metabolite profiles. In addition to metabolites commonly perturbed after IR exposure, we identified and validated a novel metabolite (hexosamine-valine-isoleucine-OH) that increased up to 150-fold after LDR and 80-fold after conventional exposures in urine. A multiplex panel consisting of hexosamine-valine-isoleucine-OH with other urinary metabolites (N6,N6,N6-trimethyllysine, carnitine, 1-methylnicotinamide, α-ketoglutaric acid) achieved robust classification performance using receiver operating characteristic curve analysis irrespective of dose rate or sex. These results show that in terms of biodosimetry, dysregulated energy metabolism is associated with IR exposure for both LDR and conventional IR exposures. These mass spectrometry data have been deposited to the NIH data repository via Metabolomics Workbench with study ID ST001790, ST001791, ST001792, ST001793, ST001806.

Graphical Abstract

1. Introduction

Threats from large-scale radiological events necessitate development of radiation countermeasures, including rapid high-throughput biodosimetry.^{1, 2} Complex exposures from a potential groundburst improvised nuclear device will include both higher dose rate and low-dose rate (LDR) ionizing radiation (IR) exposures from gamma and neutron components.³ LDR IR exposure will consist of external contamination (groundshine) and radionuclide ingestion or inhalation, of particular concern being ¹³⁷Cs. In addition to malicious radiological attacks, environmental deposition of ¹³⁷Cs is a primary concern in the event of nuclear reactor accidents⁴, such as the Fukushima Daiichi⁵ and Chernobyl⁶ sites, as well as improperly stored nuclear waste. A large proportion of past studies determining biomarker responses to IR have been conducted using x-ray or gamma radiation at a conventional dose rate of ~1 Gy/min (dose rates reviewed here ⁷). As such, there is an ongoing need to investigate how other components of a complex exposure, here LDR IR, will affect dose reconstruction.

While several comprehensive “–omic” studies have investigated possible targets for a composite biodosimetry biomarker panel⁸, relatively few have focused on the influence that dose rate may have on qualitative or quantitative dose reconstruction, especially at higher accumulated doses. Previous studies have shown that higher accumulated doses of external LDR IR in mice can lead to perturbations of mRNA and protein levels, increased genetic mutations and carcinogenesis (8 Gy, 20 mGy/day) (reviewed here ⁹), altered urinary metabolite levels (1.1 and 4.45 Gy, 3.09 mGy/min)¹⁰ and premature senescence stress in human fibroblast cells (HUVEC 6.2 Gy, 4.1mGy/h; VH10 7.8 Gy, 5 mGy/h).¹¹ While the above studies demonstrated several biophysiological consequences of LDR IR exposure, they used a static dose rate that does not replicate the biokinetics of internally deposited ¹³⁷Cs associated with nuclear fallout. To address the variable dose rate encountered in a realistic exposure scenario from internal deposition, our group previously designed a set of experiments that used ¹³⁷CsCl(_aq) and ⁹⁰SrCl_2(aq) injections in mice and demonstrated several downstream perturbations at the transcriptomic^12–15 and metabolomic levels.^15–21 However, inherent challenges associated with this design, such as the need for specialized facilities and instrument contamination from radioactive biofluids, limits the use of its application and does not effectively model external fallout exposures.

In this study, we examined the effects of variable LDR IR on mouse urine and serum small molecule signatures using a liquid chromatography mass spectrometry (LC–MS) global metabolomic approach. Mice were placed in the VAriable Dose-rate External ¹³⁷Cs irradiatoR (VADER)²², a custom-built irradiation system designed to model LDR ¹³⁷Cs exposures from both fallout and internal emitters. Mice were then removed at a single time point within a 30 d period (1 d – 30 d) to represent different accumulated doses (1 – 9.7 Gy) at different dose rates (~1 – 0.1 Gy/day). A smaller cohort of male mice exposed to an equivalent accumulated dose at a reference dose rate (0.8 Gy/min) and females (LDR and reference dose rate exposures) were analyzed for comparison. Using the machine learning algorithm Random Forests, we were able to separate LDR IR exposed groups from a control group based on their biofluid signatures. Higher perturbation was observed in urine compared to serum at earlier time points with lower accumulated doses (i.e., 1 – 3 d, 1 – 2.8 Gy). We identified a novel radiation specific spectral feature (m/z 393.2214_1.3) with up to a 150-fold increase in urine (after LDR irradiation in males) and subsequently used high-resolution isotope pattern analysis and in silico fragmentation tree computation to guide synthesis of a novel compound (hexosamine-valine-isoleucine-OH [Hex-V-I]) for validation. Intermediate metabolites involved in energy and nicotinamide metabolism (carnitine, N6,N6,N6-trimethyllysine [TML], 1-methylnicotinamide, α-ketoglutaric acid ) were altered after exposure to both LDR (~1 Gy/day) and the reference dose rate (0.8 Gy/min) IR exposure and show promise to be incorporated into multiplex biodosimetry assays. While sex-specific and dose-rate dependent and independent effects were observed, we identified urinary metabolite panels with excellent sensitivity and specificity performance in receiver operating characteristic (ROC) analysis irrespective of these variables.

2. Materials and Methods

2.1. The variable dose-rate external ¹³⁷Cs irradiator (VADER)

The VADER was designed to deliver controlled dose rates in the range 0.1 – 1 Gy/day to a cohort of up to 15 mice.²² The VADER uses ~0.5 Ci of retired ¹³⁷Cs brachytherapy seeds that are arranged in two platters placed above and below a “mouse hotel”. The platters can be placed ~0.5 – 60 cm above and below the mouse hotel allowing implementation of time-variable dose rates. Offline dosimetry of the VADER is performed annually using a NIST traceable 10x6-6 ionization chamber (Radcal Corp., Monrovia, CA). Dose uniformity across the surface was measured using EBT3 film (Ashland, Covington, KY, USA) and the variation was 15% across the hotel. A lead and high-density concrete brick shield ensured minimal radiation doses to occupationally exposed personnel (operators) inside (< 0.1 mGy/wk) and outside the room (< 0.02 mGy/wk).

The mouse hotel consists of an acrylic box (35 x 35 x 12 cm) allowing housing of ≤ 15 mice with bedding material and food/water ad libitum. Temperature (20 – 25°C), humidity (40 – 60%), airflow and lighting were fully monitored to required animal care standards (temperature/humidity sensor, HWg HTemp, TruePath Technologies Victor, NY). Environmental controls and monitoring were integrated into the mouse hotel for easy replacement in case of radiation damage. Mice were exposed to a 12 h day/night cycle (daytime illuminance: 60 lux) and monitored in real time using a 180° fisheye ELP USB camera (Amazon: HD 1080P USB CMOS board camera module [model: ELP-USBFHD01M serials]).

The reference dose rate irradiations were performed using a Gammacell-40 ¹³⁷Cs irradiator (AECL, Ottawa, Ontario, Canada) at a dose rate of 0.8 Gy/min. Animals were placed in the irradiator for total body irradiation exposure and then placed in cages for the same time intervals to match the 1, 2, 2.8 and 4.1 Gy doses in the protracted irradiation experiment, as well as time matched sham controls. Mice were euthanized by CO₂ and blood was collected via cardiac puncture. Whole blood samples (100 – 150 μl) were added to a serum separating tube, kept at room temperature for 30 minutes, and centrifuged for 10 minutes (1,300xg, 4 °C). Spot urine and serum was flash frozen and then stored at −80°C until shipped to Georgetown University Medical Center for analysis.

All animal experiments were approved by the Columbia University Institutional Animal Care and Use Committee (IACUC; approved protocol AAQ2410) and were conducted under all relevant federal and state guidelines. Male and female 8 – 10 week old C57BL/6 mice (Charles River Laboratories, Frederick, MD, USA) were irradiated in the VADER in two randomized batches (15 mice) loaded into mouse hotels (one hotel loaded into the VADER and one in the same room as a zero-dose control). Five mice per time point were irradiated in two back-to-back runs (run 1: 2, 3, or 5 d; run 2: 5, 20, or 30 d) to a total dose of 1 Gy (1 d), 2 Gy (2 d), 2.8 Gy (3 d), 4.1 Gy (5 d), 8.8 Gy (20 d), and 9.7 Gy (30 d) (Figure S1). Time matched controls were kept in a VADER cage in the same room with the same environmental parameters as the LDR exposed mice. Spot urine and serum were flash frozen and then stored as above.

2.2. Mouse TLD dosimetry

In vivo dosimetry was performed on a mouse-by-mouse basis by injecting glass encapsulated TLD chips.²³ Mice were anesthetized with 2% isoflurane delivered in 100% oxygen for < 3 min before the implantation procedure. Encapsulated TLD rods (one per mouse) were implanted through subcutaneous injection in the dorsal neck (using 12-gauge needle coupled with a needle injector [Allflex, Irving, TX, USA]) and monitored up to 48 h to ensure health status. TLD rods were removed surgically on euthanasia and read using a Harshaw 2500 TLD reader (Thermo Fisher Scientific™, Waltham, MA, USA), using a heating profile RT to 300°C (5°C/sec), short hold at 300°C, then cooled to 50°C. Dose was reconstructed based on the integrated light yield at a temperature higher than 180°C to eliminate the low temperature, time dependent, glow peak.²³

2.3. Metabolite Extraction and LC-MS Analysis

Reagents were Optima grade™, (Fisher Scientific, Hanover Park, IL, USA) and analytical standards (debrisoquine sulfate, 4-nitrobenzoic acid, carnitine, 1-methylnicotinamide, α-ketoglutaric acid, trigonelline hydrochloride, DL-indole-3-lactic acid, xanthurenic acid, citric acid, Nε,Nε,Nε-trimethyllysine hydrochloride) were obtained from Sigma-Aldrich (St. Louis, MO, USA). A novel compound, 3-methyl-2-(3-methyl-2-((2,4,5-trihydroxy-6-(hydroxymethyl)tetrahydro-2H-pyran-3-yl)amino)butanamido)pentanoic acid (or hexosamine-valine-isoleucine-OH [Hex-V-I]), was synthesized by Expert Synthesis Solutions (London, ON, Canada). The structure was confirmed by ¹H-NMR, ¹³C-NMR, ¹H-¹³C-HSQC and MS and these spectra are provided in Supplementary Figures 2–5.

Samples were prepared and analyzed as previously described.^{24, 25} Briefly, serum (5 μl) was deproteinized (195 μl 66% cold acetonitrile [ACN]) with internal standards (2 μM debrisoquine [M+H]⁺ = 176.1188; 30 μM 4-nitrobenzoic acid [M-H]⁻ = 166.0141), vortexed, incubated on ice (10 min), and centrifuged for 10 min (max speed, 4 °C). Urine (20 μl) was deproteinized (80 μl 50% cold ACN) and prepared as above. For urine and serum 1 μl of each sample were combined for a quality control (QC) sample. Samples were injected (2 μl) into a Waters Acquity Ultra Performance Liquid Chromatography (UPLC) with a BEH C18 1.7 μm, 2.1 x 50 mm column and coupled to a Xevo® G2-S quadrupole time-of-flight (QTOF) mass spectrometer (MS) (Waters, Milford, MA, USA). Negative and positive electrospray ionization (ESI) data-independent modes were used for data acquisition with leucine enkephalin ([M+H]⁺ = 556.2771, [M-H]⁻ = 554.2615) as Lock-Spray®. Operating conditions for ESI were: capillary voltage 2.75 kV, cone voltage 30 V, desolvation temperature 500°C, desolvation gas flow 1000 L/Hr. Mobile phases consisted of the following: solvent A (water/0.1% formic acid [FA]), solvent B (ACN/0.1% FA), solvent C (isopropanol [IPA]/ACN (90:10)/0.1% FA). The gradient for urine was (solvent A and B) 4.0 min 5% B, 4.0 min 20% B, 5.1 min 95% B, and 1.9 min 5% B at a flow rate of 0.5 ml/min, column temp 40 °C. The gradient for serum was (solvent A, B, and C) 4.0 min 98:2 A:B, 4.0 min 40:60 A:B, 1.5 min 2:98 A:B, 2.0 min 2:98 A:C, 0.5 min 50:50 A:C, and 1.0 min 98:2 A:B at a flow rate of 0.5 ml/min, column temp 60 °C. Blanks and QC samples were run after every 10 samples for urine and after every 5 samples for serum.

2.4. Data Processing, statistical analysis, and compound validation

Raw data files were inspected in MassLynx v.4.1 (Waters, Milford, MA, USA) before deconvolution and peak alignment in Progenesis QI (Nonlinear Dynamics, Newcastle, UK) as previously described.²⁶ Pre-processed data matrices were normalized to debrisoquine sulfate for ESI⁺ or 4-nitrobenzoic acid for ESI⁻. Multi-dimensional scaling (MDS) plots were generated for the 100 top ranked ions in ESI⁺ or ESI⁻ modes using the machine-learning algorithm Random Forests (RF) programmed in R v.2.15.2.^27,28 For multivariate analysis, a separate cohort of sham-exposed mice was used for a control. Spectral features were mined using the software MetaboLyzer, where Welch’s t-tests are used to identify statistically significant (P < 0.05) spectral features present ≥70% in both groups or a Barnard’s test for features present < 70% in a single group.²⁹ For univariate analysis, control samples consisted of pre-exposure samples from each individual mouse (urine) or from a separate cohort of sham-exposed mice (serum). Putative identification of the spectral features was determined by comparing monoisotopic mass (±10 ppm error) to the Human Metabolome Database (HMDB)³⁰ and Chemical Entities of Biological Interest database (ChEBI).³¹ Compounds were validated to the Metabolomics Standards Initiative (MSI) confidence level 1 by comparing retention time, precursor m/z, and tandem MS spectra (5 – 50 V ramping collision energy) to pure analytical standards.^{32, 33} For Hex-V-I (m/z 393.2214_1.3), we first generated a tandem MS spectrum then utilized the Sirius 4.0 with CSI:FingerID³⁴ with PubChem³⁵, HMDB, and ChemSpider³⁶ databases and NIST17 hybrid-search³⁷ programs for putative elemental formula, structural prediction with high-resolution isotope pattern analysis, and in silico fragmentation tree computation to guide synthesis of the pure standard. To determine a relative concentration range of Hex-V-I in 2 d LDR male and female urine, samples were quantified using a five-point standard curve (1 – 10,000 ng/ml). The software MetaboAnalyst 4.0 was used to generate LDR vs. reference dose rate exposure heatmaps and determine the specificity and sensitivity of markers by calculating the area under the curve (AUC) values from ROC analysis with a Random Forests (RF) classification method for combined metabolites.³⁸ Validated compounds were checked for outliers (ROUT Q=1%) and plotted in GraphPad Prism 6 (GraphPad Software, La Jolla, CA, USA). Due to a high number of zero values for Hex-V-I in control samples, the controls were grouped and were then analyzed with an ordinary one-way ANOVA in GraphPad Prism 6.

3. Results

3.1. Low-dose Rate Effects

3.1.1. Urine

We first used a RF machine learning algorithm to assess separation of groups exposed to LDR IR to a control group (visualized as an MDS plot) and a heatmap was generated of the primary (top 100 ranked) spectral features contributing to separation. For urine, separation of LDR IR exposed groups was similar to our previous ¹³⁷Cs internal emitter study.²⁰ All groups, with the exception of 1 d (1 Gy), separated approximately equal from the control along dimension 1 (classification accuracy 92%) (Figure 1A). Mice exposed to 1 Gy LDR IR showed the least variation compared to the control group, with one individual grouped with the 20 d (8.8 Gy) exposure group. Exposure groups 2, 3, 5, and 30 d (2, 2.8, 4.1, and 9.7 Gy) all showed similar variation from the control. Inspection of the heatmap indicates both lower (orange inserts) and higher concentrations (green insert) of feature subgroups across exposure groups (Figure 1B). Black inserts highlight a group of features that were similar at 1 and 20 d (1 and 8.8 Gy) to the control group contributing to their separation from other exposure groups.

Figure 1. LDR Effects Urine:

A) Multidimensional scaling plot generated from machine learning algorithm Random Forests for the positive mode data matrix in urine (classification accuracy 92%). Separation from the control group is similar for 2 – 30 d with less separation for 1 d. B) Heatmap shows several features were found in higher (green insert) or lower (orange inserts) concentration across all exposure levels. Another group of features were found at similar concentration to the control group for both 1 and 20 d (black inserts), which may explain the closer proximity of these groups to other exposure levels.

Subsequent Welch’s t-tests were used to compare post-irradiation samples to each individual’s pre-irradiation sample. Higher urinary concentrations of carnitine at 2 d (P = 0.009, FC = 2.3), 3 d (P = 0.029, FC = 1.9), 5 d (P = 0.014, FC = 1.7) and 1-methylnicotinamide at 1 d (P = 0.041, FC = 1.7), 5 d (P = 0.030, FC = 1.7) were observed after LDR IR exposure (Table 1, Figure 2). Additionally, we found changes in concentrations of TML at 2 d (P = 0.035, FC = 1.4) and α-ketoglutaric acid 2 d (P = 0.009, FC = 0.7).

Table 1.

Validated metabolites in mouse biofluids after exposure to low-dose rate ionizing radiation or a reference dose rate.

Biofluid	Metabolite	Adduct	RT min	Experimental m/z	Calculated m/z	Mass error ppm	Formula	HMDB ID	MS/MS Fragments
									Fragment 1	Fragment 1	Fragment 1
Urine	TML	H+	0.25	189.1603	189.1603	0.1	C9H20N2O2	0001325	130.0861	84.0805	60.0890
	Carnitine	H+	0.28	162.1132	162.1130	1.2	C7H15NO3	0000062	103.0393	102.0898	85.0288
	1-Methylnicotinamide	H+	0.28	137.0717	137.0715	1.5	C7H8N2O	0000699	94.0629	92.0484	78.0341
	Hex-V-I	H+	1.31	393.2214	393.2234	5.8	C17H32N2O8	-	309.1832	150.0916	98.0661
	α-Ketoglutaric acid	H−	0.31	145.0133	145.0137	2.8	C5H6O5	0000208	101.0235	73.0286	57.0333
	Citric acid	H−	0.29	191.0189	191.0192	1.7	C6H8O7	0000094	111.0084	87.0082	85.0282
	Xanthurenic acid	H+	0.85	206.0457	206.0453	1.8	C10H7NO4	0000801	188.0342	178.0507	160.0398
Serum	Trigonelline	H+	0.29	138.0551	138.0555	2.9	C7H7NO2	0000875	110.0607	94.0649	92.0499
	Indolelactic acid	H−	2.38	204.0662	204.0661	0.6	C11H11NO3	0000671	186.0529	158.0616	116.0523

Figure 2.

Biofluid concentrations of perturbed metabolites in mice exposed to low-dose rate (LDR) ionizing radiation (IR) for 1 (1 Gy), 2 (2 Gy), 3 (2.8 Gy), 5 (4.1 Gy), 20 (8.8 Gy), or 30 d (9.7 Gy). For urine, spot urine was collected from each individual mouse for comparison to its respective post-LDR IR sample. Some of the largest changes were observed in carnitine and 1-methylnicotinamide with minor changes observed in N6,N6,N6-Trimethyllysine (TML) and α-ketoglutaric acid. For serum, samples were collected from a sham exposure group of mice housed in the VADER and used for a control. Serum metabolites exhibited a more muted response to LDR IR exposure compared to urine.

(* P ≤ 0.05, White bars represent pre-irradiation levels and black bars represent post-irradiation levels, Mean ± S.E., P values determined by Welch’s t-test)

One spectral feature that was ranked as high importance in RF analysis after exposure to LDR IR in males and females was m/z 393.2214_1.3 (Male P < 0.001, 2 d [FC = 151.1], 3 d [FC = 52.9], 5 d [FC = 77.3]; Female P < 0.001, 2 d [FC = 4.2], 3 d [FC = 2.3]) (Table 1, Figure 3). This feature was additionally increased after the reference IR exposures in all mice (Figure S6). A compound with high similarity to the experimental tandem MS spectra was identified as beta-L-fructopyranosyl-L-val-L-ile-OH (PubChem CID: 101407470), however, no structures available in chemical databases met the criteria to explain the observed fragmentation pattern. A putative structure de novo was developed, Hex-V-I, based on the experimental fragmentation pattern and a pure standard was synthesized then confirmed to a MSI level 1 by matching retention time and tandem MS fragmentation, which is provided in Supplemental Figures S7–8 and Table S1. The presence of α- and β- anomers in both the standard and urine may indicate both a glucosamine and galactosamine mixture. Interestingly, the tandem MS spectrum for Hex-V-I was similar to a spectral feature previously reported in human urine using a HILIC separation coupled to a Thermo Q Exactive MS and assigned to a MSI level 3 identification.³⁹

Figure 3.

Urine concentrations of hexosamine-valine-isoleucine-OH (Hex-V-I) in male and female mice exposed to low-dose rate ionizing radiation. Control samples were grouped. At 2 d, the relative concentration range for male mice was 5.1 – 13.4 ng/ml and for female mice 7.1 – 38.5 ng/ml.

(* P ≤ 0.05, ** p ≤ 0.01, *** P ≤ 0.001, Mean ± S.E., NA – Female samples were not collected for these days. P values determined by an ANOVA with multiple comparisons)

Metabolite responses were compared to male mice from a previous study²⁰ with internal LDR ¹³⁷Cs IR at 2 – 30 d. Despite differences in exposure (¹³⁷CsCl_(aq) injections vs. external exposure) similar fold changes in both xanthurenic acid and citric acid were observed during 2 – 30 d suggesting some similarities between both internal emitters and external exposure from fallout following a radiological emergency (Figure S9).

3.1.2. Serum

Similarly, serum profiles showed separation between the exposed and control groups, however, a distinct difference between lower accumulated doses (1 – 2.8 Gy) and higher accumulated doses was observed (4.1 – 9.7 Gy) (classification accuracy 73%) (Figure S10). The purple insert of the corresponding heatmap shows a high proportion of features were unchanged from the control at 1 – 3 d (1 – 2.8 Gy) but increased in concentration from 5 – 30 d (4.1 – 9.7 Gy), and conversely, the green insert shows a smaller group of features that were lower in concentration only at 5 – 30 d. However, some features were present in higher concentration in all exposed groups (black insert). Lower concentrations of trigonelline at 1 d (P = 0.028, FC = 0.4), 3 d (P = 0.015, FC = 0.5), 5 d (P = 0.003, FC = 0.4) and indolelactic acid at 2 d (P = 0.041, FC = 0.4) were observed (Table 1, Figure 2).

3.2. Comparison of LDR, Reference Dose Rate, and Sex

3.2.1. LDR vs. Reference Dose Rate Urine

We further explored the effect of dose rate on metabolic changes in the first 5 d in urine. The reference dose rate exposure showed increased separation from the control samples at 2 – 3 d but returned closer to control levels at 5 d (classification accuracy 81%) (Figure S11), which has been previously observed in NHP urine after a total dose of 4 Gy at 0.6 Gy/min.⁴⁰ Higher separation from the control group was observed for urine signatures at 2 – 3 d after LDR exposure compared to serum. However, we observed a closer grouping to reference dose rate exposures at a higher accumulated dose from persistent LDR IR exposure (5 d, 4.1 Gy).

3.2.2. LDR vs. Reference Dose Rate Serum

Serum small molecule signatures clearly separated from the control group at 2 – 5 d after reference exposures (classification accuracy 70%) (Figure 4). After the LDR exposure, little separation from the control group was observed at earlier time points (2 – 3 d), however, at higher accumulated doses (5 d, 4.1 Gy) the effect of dose rate diminished. These results reiterate that there are clear dose rate dependent and independent effects on metabolism.

Figure 4. LDR vs. Reference Dose Rate Serum:

Multidimensional scaling plot generated from machine learning algorithm Random Forests for the positive mode data matrix in serum (classification accuracy 70%). Increased variation is observed at earlier time points (2 – 3 d) in serum after irradiation with a reference dose rate (0.8 Gy/min) compared to a low-dose rate (~1 Gy/day). At 5 d less influence of dose rate is observed possibly due to a higher accumulated dose (4.1 Gy).

3.2.3. Pathway and ROC Analysis

Pathway analysis indicated metabolites involved in energy metabolism (fatty acid β oxidation and tricarboxylic acid [TCA] cycle) had similar perturbations irrespective of dose rate, which also correlates to several previous studies on the effects of IR exposure on biofluid metabolite levels (Figure 5, pathways truncated). ROC curve analysis showed that excellent sensitivity and specificity could be achieved at 2 d (2 Gy) irrespective of sex (AUC = 0.98) using a three-metabolite panel (TML, 1-methylnicotinamide, and Hex-V-I) or dose rate (AUC = 0.96 – 1.00) using a five-metabolite panel (TML, carnitine, 1-methylnicotinamide, α-ketoglutaric acid, and Hex-V-I) (Figure 6).

Figure 5. Pathway Analysis:

Several metabolites identified in murine biofluids show consistent perturbation after exposure to different doses and dose-rates of ionizing radiation. Many of these metabolites are involved in energy metabolism (fatty acid β oxidation and tricarboxylic acid cycle [pathways truncated]).

a (Mak et al., 2014, Metabolomics)⁵⁹ b (Pannkuk et al., 2020, Metabolites)⁵¹ c (Goudarzi et al., 2014, Radiat Res)²⁰ d (Goudarzi et al., 2015, Radiat Res)¹⁶ e (Goudarzi et al., 2014, Radiat Environ Biophys)¹⁰ f (Lanz et al., 2009, Radiat Res)⁶⁰ g (Laiakis et al., 2016, Mutat Res)⁶¹ h (Zhao et al., 2017, Mol Biosyst)⁴³ i (Tyagi et al., 2020, Sci Rep) ⁶²

TML - N6,N6,N6-Trimethyllysine

*These may show different trends in humans or nonhuman primate models

Figure 6. ROC Analysis:

Receiver operating characteristic (ROC) curves for urinary metabolites that show excellent sensitivity and specificity at 2 d (2 Gy) irrespective of sex (N6,N6,N6-Trimethyllysine, 1-methylnicotinamide, and hexosamine-valine-isoleucine-OH) or dose rate (N6,N6,N6-Trimethyllysine, carnitine, 1-methylnicotinamide, α-ketoglutaric acid, and hexosamine-valine-isoleucine-OH).

4. Discussion

In this study, we utilized small molecule profiling on biofluids from mice exposed to variable LDR IR exposures over a 30 d time span to assess its utility in biodosimetry following fallout-type exposures. We first assessed the ability to identify individuals exposed to LDR IR. By identifying urine and serum metabolite profiles with the use of machine learning algorithms, we successfully separated individuals exposed to LDR IR from a control group. Secondly, we explored how variables including dose rate and sex could influence small molecule signatures. While this study reiterates the presence of dose rate dependent effects, there is also a need for markers independent of dose rate or sex to simplify small molecule biodosimetry assays. Here, we show metabolites involved in energy metabolism may change irrespective of these variables and identify individuals with IR induced tissue damage. In addition, we have identified a novel urinary glycopeptide that greatly increased in concentration after LDR and reference exposures.

Exposure to IR can have profound effects on host energy metabolism (e.g., TCA cycle intermediates, glycolysis, fatty acid β oxidation) and perturbed metabolite levels integral to these pathways have been identified across several animal models, biofluids and tissues, and show relatively consistent trends. The TCA cycle is the primary route for acetyl-CoA oxidation. TCA cycle intermediates are more amenable to analysis by GC/MS platforms and several studies utilizing this approach provide a holistic view of how IR exposure affects this pathway.^41–43 Concentrations of urinary TCA cycle intermediates generally begin to decrease 5 h after higher dose rate IR exposures with the largest decline occurring at ~3 d. Decreased levels can remain up to 7 d; however, there may be a dose dependent effect, as some intermediates will return to pre-exposure levels at lower doses (2 – 4 Gy) compared to higher doses (6 – 10 Gy).^41–43 Although a standard untargeted LC-MS based analysis will typically not detect all TCA intermediates, citric acid and α-ketoglutaric acid are routinely identified and may serve as a proxy for perturbations of this specific pathway and several studies have decreased levels of these metabolites in urine after IR exposure.⁴⁴ Another primary pathway in energy metabolism, fatty acid β oxidation, involves the transfer of carnitine and long-chain fatty acids into the mitochondria. Posttranslational methylation of specific proteins produces TML for enzymatic conversion to carnitine. Fatty acids are oxidized after conversion to a fatty acyl-CoA (EC:6.2.1.3) and bound to carnitine to form an acylcarnitine (EC:2.3.1.21) for transfer into the mitochondrial matrix and acetyl-CoA production. Here we observed increases in urinary levels of both TML and carnitine in a mouse model after both LDR and reference IR exposure, which has also been observed in NHP models.^{40, 45} Together, consistent results across studies suggest that perturbed urinary levels of citric acid and carnitine, especially at earlier time points (e.g., 2 – 3 d) and lower doses (e.g., ~1 – 2 Gy equivalent in humans), may provide valuable markers for early triage as well as later definitive dose estimation for both LDR and reference exposures.

In addition to perturbations in energy metabolism, variation in tryptophan metabolites has been observed across multiple animal models. Free tryptophan is primarily obtained through the diet and metabolized by the host mainly through kynurenine degradation, a smaller percentage is used for serotonin or protein synthesis, or metabolized by the host microbiota. NAD⁺ production occurs via the kynurenine pathway through tryptophan metabolism to kynurenine (EC:3.5.1.9), 3-hydroxykynurenine (EC:1.14.13.9), and nicotinamide (EC:3.5.1.19). Excess kynurenine and 3-hydroxykynurenine can be transaminated into kynurenic acid (EC:2.6.1.7) or xanthurenic acid (EC:2.6.1.7). Due to enzyme kinetics, increases in kynurenic acid or xanthurenic acid will usually be associated with increased substrate concentrations, such as acute increases in available tryptophan.⁴⁶ Alternatively, tryptophan can be used for serotonin synthesis through hydroxylation to 5-hydroxytryptophan (EC:1.14.16.4) and subsequent decarboxylation (EC:4.1.1.28). Also, the host microbiota is intricately linked to tryptophan availability and metabolism in the gut and can produce indole derivatives. Indolelactic acid can be produced through indolelactic acid dehydrogenase or aromatic amino acid aminotransferase pathways by Bifidobacterium and Lactobacillus.⁴⁷ Decreases observed in xanthurenic acid and indolelactic acid^{16, 48} suggests dysregulation to tryptophan metabolism post-irradiation, however, the association with diet and host microbiota may complicate the use of these biomarkers in biodosimetry.

Changes were observed in additional metabolites (trigonelline and 1-methylnicotinamide) that have been identified in previous radiation metabolomics experiments. Trigonelline is an alkaloid derivative of nicotinic acid found in many plants that has been detected at lower levels in rat serum 1 d after a 7 Gy exposure.⁴⁹ Methylated nicotinamide (1-methylnicotinamide) has been previously identified as both higher⁵⁰ and lower⁵¹ urinary concentrations in mice after IR exposure. Methylated products of nicotinamide and nicotinic acid are interesting due to their involvement in the salvage pathway for NAD⁺ production (as discussed above), its disruption during radiotherapy⁵², and potential in biodosimetry.

The largest fold increase observed in this study was for spectral feature m/z 393.2214_1.3, which was a previously unknown metabolite. We subsequently synthesized a compound guided by the tandem MS spectra and then validated this compound as a glycopeptide consisting of a hexosamine carbohydrate bound to a valine-isoleucine dipeptide (Hex-V-I). While exposure to IR has been shown to have dramatic effects on blood protein glycosylation after localized irradiation of skin⁵³, radiotherapy⁵⁴, and on intestinal endothelial glycosylation⁵⁵, the structure of Hex-V-I is relatively simple compared to complex glycoproteins and little information on these exist in the literature beyond their chemical synthesis.⁵⁶ As unirradiated females have higher levels than males, a possible hormonal origin may contribute to the observed increases as well. Although the biophysiological origins and consequences of Hex-V-I remains to be determined, the high fold changes seen in this novel metabolite in both males and females makes this a promising marker for IR exposure.

5. Conclusions

Potential IR exposures from a nuclear detonation will include both higher dose “prompt” IR and LDR IR from radioactive fallout. Fallout will pose risks from ¹³⁷Cs dispersal and subsequent external exposures (and internal emitters if ingested or inhaled). Challenges for small molecule biodosimetry lie within inherent differences in metabolite levels across the population (e.g., genetic diversity, sex, pre-existing conditions) and different effects of dose rate and combined injury (among others) on metabolic perturbations (see ⁵⁷). Further challenges are within the recent emergence of metabolomics technology respective to other “-omic” fields, with a primary bottleneck being in the structural annotation of unknown spectral features. Despite these challenges, immense advances have been made in the field of radiation metabolomics, including identification of these inherent metabolite differences and common pathways perturbed after IR exposure. Here, we show changes occur in metabolites involved in energy pathways regardless of dose rate or sex with high sensitivity and specificity. As these pathways are vital in energy production, they are highly evolutionarily conserved and should translate well from mouse models to humans. Additionally, we have identified a novel metabolite, Hex-V-I, that shows high increases in urine after LDR exposure and serves as a target for future studies on its biophysiological function and as a possible biomarker of radiation exposure. These compounds, in addition to other biomarkers (e.g., proteins, genes, miRNA targets), may provide composite compound panels that can provide rapid triage or dose prediction irrespective of the above inherent and extrinsic differences.⁵⁸

Supplementary Material

Supp Info

Table S1. Putative fragment structures, m/z, and formulas for hexosamine-valine-isoleucine-OH.

Figure S1. Study design and sample sizes. (black triangles, male; black crosses, female)

Figure S2. ¹H-NMR spectrum of hexosamine-valine-isoleucine-OH.

Figure S3. ¹³C-NMR spectrum of hexosamine-valine-isoleucine-OH.

Figure S4. ¹H-¹³C-HSQC NMR spectrum of synthesized hexosamine-valine-isoleucine-OH.

Figure S5. ESI-TOF-MS spectrum of synthesized hexosamine-valine-isoleucine-OH in positive mode.

Figure S6. Increases in spectral feature 393.2214_1.3 in mouse urine after reference dose rate exposures.

Figure S7. UPLC chromatogram of spectral feature 393.2214_1.3 in mouse urine after IR exposure and of hexosamine-valine-isoleucine-OH standard.

Figure S8. Tandem MS spectrum in positive mode of spectral feature 393.2214_1.3 in mouse urine after LDR IR exposure (top) and of hexosamine-valine-isoleucine-OH standard (bottom).

Figure S9. Fold changes in urinary xanthurenic acid and citric acid levels after internal (¹³⁷CsCl_(aq) injection, 2 [2 Gy], 5 [4.1 Gy], 20 [9.5 Gy], or 30 d [9.9 Gy]) and external (2 [2 Gy], 5 [4.1 Gy], 20 [8.8 Gy], or 30 d [9.7 Gy]) low-dose rate (LDR) ionizing radiation exposure. Similar fold changes suggest similar perturbation in tryptophan and energy metabolism occurs from LDR IR exposure regardless of internal ingestion or fallout scenarios.

Figure S10. LDR Effects Serum: A) Multidimensional scaling (MDS) plot generated from machine learning algorithm Random Forests for the positive mode data matrix in serum (classification accuracy 73%). Separation from the control group is observed at earlier time points (1 – 3 d) along dimension 2, however, at higher accumulated doses (> 4 Gy) samples group together and show highest variation from control along dimension 1. B) Heatmap illustrating changes in spectral features contributing to the variation observed in the MDS plot. The purple box highlights that several features are increasing in concentration from 5 – 30 d and conversely the green box insert highlights features that decrease in concentration from 5 – 30 d. The black box insert highlights some features increase in concentration at all exposure levels.

Figure S11. LDR vs. Reference Dose Rate Urine: Multidimensional scaling plot generated from machine learning algorithm Random Forests for the positive mode data matrix in urine (classification accuracy 81%). Unlike serum, the reference dose rate exposure show increased separation from the control group at 2 – 3 d but returned closer to the control at 5 d. Higher separation from the control group was observed for urine signatures at 2 – 3 d after low-dose rate (LDR) exposure compared to serum.