Salivary metabolomics of total body irradiated non-human primates reveals long term normal tissue responses to radiation.

Abstract

Purpose

To identify metabolomic biomarkers of acute radiation exposure in saliva that show time-dependent changes.

Methods and Materials

Non-human primates (NHPs) were exposed to 4 Gy of total body irradiation with γ-rays. Saliva was collected from seven animals twice prior to and at days 1, 3, 5, 7, 15, 21,28, and 60 after irradiation. Profiling was conducted with liquid chromatography time-of-flight mass spectrometry. Multivariate data analysis and potential biomarker identification was conducted through Random Forests and the software MetaboAnalyst. Candidate biomarkers were validated through tandem mass spectrometry (MS/MS) and receiver operating characteristic (ROC) curves were constructed to show the diagnostic ability of the signature over time.

Results and Conclusions

Untargeted metabolomic analysis revealed significant and persistent effects up to the 60 days evaluated in this study. Biomarkers spanning primarily amino acids and nucleotides were identified with a significant number of them showing long term responses. Fifteen biomarkers showed high statistical significance in the first week after irradiation, and sixteen at >7 days after irradiation [false discovery rate (FDR) adjusted p<0.05]. The combination of the biomarkers in a single biosignature was able to accurately show the diagnostic ability of the signature in a binary classifier system with ROC curves. Radiation therefore can alter the metabolome in saliva and metabolomics could effectively be used to monitor radiation responses, as a biodosimetry method, in the event of a radiological incident; saliva metabolomics also has potential relevance in a clinical setting.

Summary

The purpose of this study was to evaluate changes in the salivary metabolome associated with total body radiation exposure with the use of untargeted metabolomics. Biomarkers were identified, primarily amino acids and nucleotides, with a significant number showing long term responses (up to 60 days after irradiation). Radiation therefore can alter the metabolome in saliva and salivary biomarkers could be used to monitor radiation responses.

Introduction

During a radiological event or accidental exposure, thousands of individuals may receive significant doses of ionizing radiation (IR) that can prove to be detrimental for radiosensitive tissues. Multiple reviews have summarized the ongoing efforts to develop biodosimetric methods based on easily accessible biofluids and tissues for the rapid identification of exposed populations such that quick medical management can be provided to those in need^{17, 24, 25, 28}. For this, saliva is an attractive biofluid for analysis as it is easily accessible, contains biomolecules that are reflective of radiation-induced normal tissue injury and can be mined serially for repeat monitoring of radiation responses in a qualitative and quantitative manner. Saliva has been previously used for screening and diagnostic purposes, particularly for cancer patients and those with autoimmune diseases^{8, 15, 16, 22}. Recent efforts in radiation biology have utilized saliva for detection of radiation induced changes ^{11, 12}. Both metabolites and protein biomarkers showed altered levels following acute exposures in mice or dose escalation in head and neck cancer patients, although saliva itself can also contain other constituents such as DNA and bacterial or viral products ³².

Salivary glands and the oral mucosa are considered highly radiosensitive tissues ^{6, 19}. In addiction to accidental exposures, during conventional radiotherapy, particularly for head and neck cancers, the normal tissue and salivary glands are exposed to high cumulative doses that can lead to injury and structural changes of the tissue, but also result in depletion of salivary gland stem cells ¹. Similar responses with loss of function and apoptosis of acinar cells have been documented in mice and nonhuman primates (NHPs) following a single acute exposure ⁶. High fractionated doses or acute high dose exposures, can lead to xerostomia with eventual dysphagia that can adversely affect the quality of life of a patient, further leading to nutritional loss from decreased appetite, reduced caloric intake and even possible development of mucositis ^{1, 6, 25}. In cases where parotid glandular stem cells are not depleted, recovery of adequate salivary production can take up to a year or more after radiation treatment ²⁹ Various treatments have been considered for the management of radiation induced adverse effects, from administration of radioprotectors, to stem cells and tissue organoids ^{1, 13, 20}

Here we investigated long term responses to an acute exposure of 4 Gy of γ radiation in an NHP animal model to determine the potential of using salivary biomarkers as an indicator of radiation exposure and longitudinal effects. We employed liquid chromatography time-of-flight mass spectrometry (LC-TOFMS) techniques in combination with metabolomic computational methods to develop a biosignature that could distinguish pre-exposure vs. post-exposure conditions. Saliva samples were analyzed up to 60 days after total body irradiation in order to monitor for the potential of persistent metabolic changes reflected in this biofluid.

Materials and Methods

Chemicals

All chemicals were of the highest purity and all reagents were of LC-MS grade. The chemicals that were used for positive identification of the candidate biomarkers through MS/MS were acquired from Sigma-Aldrich (St. Louis, MO). Pseudouridine was acquired from MP Biomedicals. Sebacic acid was acquired from Fisher Scientific (Grand Island, NY).

Animal Irradiations

In vivo activities were conducted at an AAALAC accredited facility and the experimental protocol was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC). Three male and four female Rhesus (Macaca mulatta) with a range of approximately 5-7 years of age (body weights ranging from 6.7-8.9 kg for males and 4.8-6.2 kg for females) were of Chinese origin. Animals were housed in a controlled environment (targeted ranges: temperature 21 ± 3°C; relative humidity 50 ± 20%; 12 hours light, 12 hours dark). A standard certified commercial chow (ENVIGO Teklad Certified Hi-Fiber Primate Diet #7195C) was provided twice daily. Treats or fresh fruits/vegetables or juice (no citrus) were provided as part of the animal enrichment program. Supplemental nutritional support was provided when animals became anorexic and when body weight decreased by >10% from baseline. Municipal tap water was provided ad libitum. Supportive care was provided on an as-needed basis and by recommendation of the veterinarian. Body weights, body temperature, food appetence, cage side clinical sign checks, and hematology data were collected throughout the study.

Prior to irradiation day, the radiation dosimetry was conducted using an acrylic phantom. Animal body dimensions were measured before exposure and were used to calculate exposure time for each individual animal. Animals were acclimated on 3 occasions to the exposure chair prior to irradiation. Animals were fasted overnight and administered ondansetron (2 mg/mL, 1.5 mg/kg, IM) prior to irradiation (between 45–90 minutes) and following irradiation (between 30-45 minutes) to manage potential emesis. Unanesthetized animals were exposed in a vertical position to a total body dose of 4 Gy γ radiation from a ⁶⁰Co source using a clinical unit (Theratron 1000) at a dose rate of approximately 60 cGy/min. To provide a homogenous dose distribution, animals were initially irradiated in the antero-posterior (AP) position for half of the exposure and then turned for the second half of the exposure in the postero-anterior position for a total body dose of 4 Gy. A Farmer ionization chamber was used as the primary dosimetry methodology. The mean dose given to the group was 3.87 Gy. As a secondary dose verification, two dosimeters (Landauer, Inc. Model nanoDots^™) were also placed on each animal under a gel bolus build-up of ~5 mm (superflab). Total irradiation time was approximately 7 minutes.

The 4 Gy radiation dose administered was sublethal and below a typical LD_50/60 total body dose of ~6 Gy. A comparable dose in humans is ~2.2 Gy, with an LD_50/60 in humans of ~4.5 Gy ¹⁸. Most animals had no major changes in feces appearance or diarrhea, no change in activity level or posture. One male animal (ID#2004), however, had moderate to severe clinical signs observed mostly during weeks 2–3 post irradiation including decreased appetite and decreased activity. This animal was noted to have buccal ulceration on D17 and gastric ulcers for which supportive care was administered.

Saliva sample collection and processing

Saliva samples were obtained from animals when possible on Days −8 and −3 pre irradiation and on Days 1, 3, 5, 7, 15, 21, 28, and 60 after irradiation. Immediately prior to collection, animals were restrained in a chair with their arms and head secured. A bite stick (or equivalent) was inserted into the mouth to keep the animal’s mouth open and a cotton rope (SalivaBio oral swab) was inserted in the oral cavity to adsorb the maximum volume of saliva. The cotton rope was centrifuged in the swab storage tube for 15 minutes (1,500 xg at 4°C) to extract the saliva. Blood and/or particles in saliva were documented. Saliva was aliquoted and stored at −70°C until shipment.

Samples were thawed on ice and diluted in a 1:1 v/v with 50%:50% acetonitrile:water containing internal standards (5 μM chlorpropamide, 30 μM 4-nitrobenzoic acid, 4 μM debrisoquine sulfate), followed by a centrifugation step at 13,000 g for 20 min at 4°C. The supernatant was transferred to a mass spectrometry vial. A quality control (QC) sample was prepared to monitor for chromatographic integrity, including retention time shifts. Two μL of each sample were injected into an Ultra Performance Liquid Chromatography system (UPLC) coupled to a Xevo G2S TOFMS. Data acquisition was conducted in MS^E function and mass correction was achieved by intermittent injections of leucine enkephalin as Lockspray^™. The solvents consisted of A (water + 0.1 % formic acid) and B (acetonitrile + 0.1 % formic acid) and the gradient and column were described previously ¹².

Data analysis and validation

Deconvolution was conducted with the software Progenesis QI (NonLinear Dynamics, Newcastle UK). Peak alignment was based on a QC chromatogram chosen by the software. Deconvoluted data was normalized to total protein levels, quantitated with the microBCA^™ assay (Thermo Fisher Scientific Inc, Grand Island NY). Data was split into two categories, data points in the first week after irradiation compared to pre-irradiation and data points >15 days compared to pre-irradiation. Normalized data (ESI⁺ and ESI⁻) was initially analyzed with the machine learning algorithm Random Forests (RF) ³ in the programming language R and multidimensional scaling plots (MDS) were constructed from the top 100 ranked ions. MetaboAnalyst ⁴ was employed to identify statistically significant ions through a one-way ANOVA test with Tukey’s post hoc test and false discovery rate correction (FDR) of 0.05. Putative IDs of statistically significant ions were assigned through the databases Human Metabolome Database (HMDB) ³⁰, Kyoto Encyclopedia of Genes and Genomes (KEGG) ¹⁰, and LIPID MAPS® ²⁷ with a ppm error of 10.

Validation of candidate biomarkers was conducted through MS/MS with ramping collision energy. Fragmentation patterns of the candidate biomarkers in a QC sample were matched to those from pure chemicals and cross-referenced to the database METLIN ²³. Pathway enrichment and heatmap generation was also conducted through MetaboAnalyst based on the positively identified metabolites. The heatmap was constructed with Euclidean distance and the Ward clustering algorithm. Data was graphed with the software Prism 6 (GraphPad Software Inc., LaJolla CA) after outlier identification and removal with the ROUT method and Q set to 1 % and represented as mean± standard error of the mean (SEM). Only metabolites that had similar levels in Pre-8 and Pre-3 were considered as biomarkers of radiation exposure. Statistical analysis was conducted through one-way ANOVA with Tukey’s as a multiple comparison test (all time points compared to −3 days). A p-value of <0.05 was considered statistically significant. ROC curves were also constructed through MetaboAnalyst with RF selected as the classification and feature ranking method. All models included an area under the curve (AUC) value and depiction of the 95% confidence band. An AUC >0.9 was considered excellent, while >0.8 was considered a good model.

Results

Following irradiation, hematological parameters such as white blood cells, neutrophils, platelets, and red blood cells decreased, providing evidence of radiation-induced myelosuppression with pancytopenia (data not shown), indicating some normal tissue toxicity due to radiation exposure. Protein quantification in saliva samples showed no statistical significance between the groups (p=0.234), therefore each ion was normalized to each sample’s respective protein level. Initial analysis to evaluate overall changes in metabolic profiles in a time dependent manner was conducted with RF. One sample from the D1 cohort was removed from the analysis, as its metabolic profile was an outlier. This was the animal that presented with moderate to severe clinical symptoms at a later time point (ID#2004). Construction of an MDS plot (top 100 ranked ions, ESI⁻ data) showed the existence of a time dependent alteration of the metabolome, with changes persisting up to 60 days after-IR (Figure 1). The classification accuracy was 78.7 %. Exploring the profiles in the first week and later time points (>2 weeks after IR), the classification accuracy increased to 97.1% and 97.2% respectively.

Figure 1:

RF analysis of the ESI⁻ data as a whole and separated by first week or later time points. MDS plots of the top 100 ranked ions allows for separation group clustering according to their unique metabolic profiles.

Twenty-one metabolites were validated through tandem mass spectrometry (Table 1). Metabolites spanned both ESI⁺ and ESI⁻. Guanosine was significant as an [M+H]⁺ and [M+Na]⁺ adduct. The MS/MS profiles of L-leucine and isoleucine could not be distinguished and therefore the identity of this metabolite is presented as a combination. The same applies to uridine and pseudouridine. Overall 15 biomarkers exhibited statistically significant changes (p<0.05) in the first week after IR, while 16 biomarkers at the later time points (Table 1). Eleven metabolites were common for the two time points and therefore, represent stable and persistent biomarkers after radiation injury. Twelve representative metabolites are shown in Figure 2. L-Citrulline, guanosine, propionylcarnitine, and uridine/pseudouridine showed an initial sharp spike at one day after IR, returning close to normal levels. On the other hand, adenine, 4-pyridoxic acid, L-tyrosine, glutaric acid, pipecolic acid, and leucine/isoleucine showed a progressive decrease over time. Glutamic acid remained unchanged until D60, when it sharply increased in saliva. Xanthine overall showed changes, however each separate time point was not statistically significant compared to pre-IR levels. Nicotinamide (niacinamide) was initially identified as a biomarker, however the p-value was not statistically significant following FDR correction. The individual normalized abundance of each biomarker for each animal over time are presented in Supplementary Table 1, which also includes the data for animal #2004 that was removed from the overall analysis.

Table 1:

Validated salivary biomarkers

					≤7 days	>7 days
	m/z	Ret. Time (min)	Adduct	Validated Metabolite	p-value FDR corrected	p-value FDR corrected
1	113.0356	0.61	[M+H]+	Uracil	0.0042	0.0035
2	175.12	0.43	[M+H]+	L-Arginine	<0.0001	0.0067
3	198.0864	0.52	[M+Na]+	L-Citrulline	<0.0001	0.1435
4	218.14	0.86	[M+H]+	Propionylcarnitine	<0.0001	0.2719
5	267.0602	0.65	[M+Na]+	Uridine/Pseudouridine	<0.0001	0.0594
6	284.1005	0.7	[M+H]+	Guanosine	0.0016	0.4457
7	306.0826	0.68	[M+Na]+	Guanosine	0.0096	0.0155
8	130.0873	0.43	[M+H]+	Pipecolic acid	0.0009	0.0124
9	123.0563	0.55	[M+H]+	Nicotinamide (Niacinamide)	0.0694	0.1588
10	136.0606	0.75	[M+H]+	Adenine	0.0006	0.0001
11	130.0867	0.8	[M-H]−	L-Leucine/Isoleucine	0.0001	<0.0001
12	131.0345	1.01	[M-H]−	Glutaric acid	0.1851	<0.0001
13	135.0304	0.6	[M-H]−	Hypoxanthine	0.2645	0.0007
14	146.0453	0.5	[M-H]−	Glutamic acid	0.2921	0.0001
15	151.0255	0.64	[M-H]−	Xanthine	0.408	0.0027
16	180.0657	0.7	[M-H]−	L-Tyrosine	<0.0001	0.0002
17	182.045	0.8	[M-H]−	4-Pyridoxic acid	0.0227	0.0059
18	201.1123	0.43	[M-H]−	Sebacic acid	0.005	0.0012
19	257.1749	7.31	[M-H]−	Tetradecanedioic acid	0.758	0.0012
20	154.0615	0.44	[M-H]−	L-Histidine	0.025	0.0035
21	164.0709	1.15	[M-H]−	L-Phenylalanine	0.0109	0.029

Figure 2:

Select metabolites from pre-irradiation to 60 days after IR. Metabolites were chosen to reflect stable levels in the week before irradiation. Light grey highlights statistically significant changes in the first week after-IR, while darker grey at the later time points. All asterisks represent a p-value of <0.05. Data are presented as mean ± SEM. Dotted lines are for reference to pre-IR samples.

The general changes of the identified metabolites are shown as an average of each group in a heatmap in Figure 3A. Three groups of profiles were immediately evident, a group that showed reduced levels following radiation exposure, a second that was immediately increased at D1 after IR, and a third that was upregulated at the later time points. Pathway enrichment based on the identified biomarkers revealed associations with amino acid metabolic pathways, purine metabolism, and urea cycle as some of the most prominent dysregulated pathways.

Figure 3:

A) A heatmap representation of the long term values of the identified metabolites. Values are averaged for each group. B) Metabolic pathway enrichment based on the identified metabolites to identify pathway specific alterations due to radiation exposure.

Based on the statistically significant metabolites, two signatures were developed spanning the first week and the later time points after IR with some overlap of biomarkers (Table 1). These signatures were used to construct ROC curves and evaluate the sensitivity and specificity. Shown in Figure 4A, a model utilizing 15 biomarkers gave AUC values close to or exceeding 0.9. The same applied for the later time points with the model utilizing 16 biomarkers, reaching a value close to 1 at day 60 (Figure 4B). Reanalyzing the data by combining all the biomarkers in a single model also led to comparable results with AUC’s for each time point close to or >0.9 (Supplementary Figure 1).

Figure 4:

ROC curves for the first week after-IR based on a metabolic signature of 15 metabolites and >15 days after-IR based on a metabolic signature of 16 metabolites, as presented in Table 1. An AUC of >0.9 represents an excellent classification model, while 0.8-0.9 fairs as good. Grey areas signify the confidence intervals.

Discussion

In this study, we employed untargeted metabolomics to investigate changes in salivary small molecules as a result of total body radiation exposure and we identified long term responses from an acute exposure. Metabolomics can provide a snapshot of the overall status of an organism or a tissue in a rapid and efficient way. NHPs were exposed to 4 Gy, corresponding to ~2.2 Gy in humans, with no shielding and samples were acquired prior to and up to 60 days after IR in order to investigate the acute effects of radiation on normal tissue after a single exposure. Untargeted metabolomics showed that changes in the overall metabolome persisted overtime and were distinct from the pre-IR samples (Figure 1). The dose used in this study results in apoptosis of acinar cells ⁶, although it was substantially much lower than the total accumulated dose received during radiotherapy. However, it is evident through the analysis that changes are immediate and long lasting even with a relatively low dose, by radiotherapy standards. Nonetheless, the dose used is relevant during a radiological event, whether that is accidental or intentional, and therefore the use of saliva is a promising biofluid for biodosimetry. The radiation signature identified provides evidence that it is possible to construct a targeted assay in the future that will not only differentiate between exposed and unexposed individuals, but also monitor for radiation related damage with high sensitivity and specificity.

Furthermore, following biomarker identification, targeted metabolomics can quantitate such metabolites with high sensitivity as to be used in a clinical setting, already considered for inborn errors of metabolism ^{5, 9}. Saliva is an attractive and promising biofluid as its acquisition is non-invasive and its contents have been characterized as significantly enriched in various disease states, e.g. cancer^{8, 11, 22, 31}. Radiation exposure, whether accidental or as part of medical treatment, can have severe and lasting consequences in the functionality of the salivary glands. Individuals exposed to high doses, with the head in the field of exposure, can suffer from xerostomia, alteration of the mucosa with gradual development of mucositis, fibrosis, changes in the oral cavity flora, and reduced taste ^{2, 6, 25}. The loss of function and apoptosis of acinar cells can show a limited reversal over time, however the immediate radiation effects can result not only in a reduced quality of life but also in dental caries, chronic infections of the oral cavity, and periodontitis ^{6, 25}. For physicians, it is important to recognize early signs of tissue injury or increased individual sensitivity and intervene to protect the tissue or modify treatment regimens. For identification of radiation exposure, saliva can provide a plethora of biomarkers that can be mined to create radiation signatures for the purposes of radiation biodosimetry.

The metabolites identified in this study primarily belong in the categories of amino acids and nucleotides. As expected and previously seen with other types of analyses ¹⁷, there was an initial and sharp response in the first 24 h after irradiation that subsided over time, attributed to acute radiation responses ²⁶. However, as tissue remodeling and delayed cell death through random necrosis in addition to lymphocytic infiltration ²⁶ is initiated, different sets of biomarkers were either activated, such as uridine and guanosine, or showed a progressive reduction in levels, such as adenine and L-tyrosine. Interestingly, in addition to cell death and tissue remodeling, cells can become senescent and possibly contribute to the observed changes. Marmary et al ¹⁴ previously identified that loss of salivary function was closely associated with senescent cells that secreted factors such as PAI-1 and IL-6. Although no profound apoptosis or necrosis was identified in their experimental model, they noted that pretreatment with IL-6 prevented both senescence and hypofunction of the salivary glands ¹⁴. Further research, such as with metabolomics, could potentially identify other treatments that can rescue the salivary glands and therefore increase the quality of life of exposed individuals, without resection and later implantation.

It is also important to note that radiation itself may result in changes in the microbiome of the oral cavity or that different cancer patients can have a variable microbiome in their oral cavity. Head and neck patients treated with intensity modulated radiation therapy (IMRT) with or without chemotherapy showed a shift towards opportunistic bacteria ²¹. On the other hand, Human Papilloma Virus (HPV) positive or negative patients with head and neck squamous cell carcinoma showed differential outcomes to radiotherapy based on their oral microbiome ⁷. It is possible therefore that some of the metabolomic biomarkers identified in this study may be due to changes in the oral microbiome, rendering important follow-up studies that will focus on such effects of radiation on the oral microbiome and its metabolomic profiling. Changes in the oral microbiome following radiation exposure in healthy individuals would be informative, however these can only be conducted in animal models with the doses relevant to the scenarios of biodosimetry or the clinic.

Our study, to the best of our knowledge, is the first to look at salivary metabolomics in response to acute radiation exposures and effects from normal tissue responses in this biofluid in NHPs. However, some limitations are present that include the lack of an independent sham irradiated cohort that can account for possible stress related changes to metabolism due to restraint during irradiation, or supplementation in select cases for nutritional support or with parenteral fluids, which may have the potential to alter metabolism. Long term follow-up studies will determine whether such effects are persistent and detectable and can be transferred to a more clinically relevant scenario. Additionally, untargeted metabolomics is still technically laborious, however identification of biomarkers can lead to development of targeted assays that are becoming more widely utilized in clinical laboratories, such as in the case of inborn errors of metabolism. Finally, further correlative studies with clinical parameters, such as leukopenia or development of oral indicators of tissue toxicity, may allow for the development of predictive panels prior to clinical symptoms.