Early molecular markers for retrospective biodosimetry and prediction of acute health effects

Combination of terms 'gene expression and radiation and dosimetry' generated 2523 entries in PubMed (03/2021) of which 122 were selected for further examination, leaving finally 72 publications (based on the abstract) for detailed analysis. Among them 45 publications were chosen for extracting exposure and methodological details to develop table 1 which aggregates detailed description of these publications shown in supplemental table 1 (available online at stacks.iop.org/JRP/42/010503/mmedia). These citations are provided in a separate chapter of the references for convenience reason to the reader. This review does not cover all relevant articles, but it certainly provides an overview and important features to this topic.

Table 1. Summary on most frequently examined candidate genes for retrospective dosimetry and H-ARS (hematological acute radiation syndrome) prediction. The table is ordered per gene and each row represents another radiation exposure quality (e.g. publications using x-ray, γ-ray, neutron or α-exposures). The last column depicts the corresponding number of publications followed by references in parenthesis and depicted in a separate chapter of the manuscript references. Abbreviations: FC, Fold change relative to an unexposed reference, PPV, positive predictive value; NPV, negative predictive value; ROC, receiver operator characteristic curve. The references here can be found in a supplementary reference list.

Applying radiation-induced GE changes in the peripheral blood for biodosimetry purposes employing microarrays as well as quantitative RT-PCR (qRT-PCR), was independently recognized by several groups around 2000 [45–50]. Earlier work (⩽2002) examined applicability of proto-oncogenes like c-fos, c-jun, c-myc and c-Ha-ras [45, 51] at exposures ⩽ 2 Gy. Soon, other gene targets belonging to the TP53 signal transduction pathway (e.g. GADD45) involved in DNA-repair, cell cycle control (e.g. CDKN1A) and cell death (e.g. BAX, TNFSF9) and other pathways e.g. immunological response (e.g. IFN-gamma) replaced these so called 'immediate early (proto-onco) genes' [50–53]. GADD45 (Growth arrest and DNA damage gene [54]) was one of the earliest TP53 related genes and many more were examined in the following decades [50].

Beside this gnostic approach, microarrays and bioinformatic analysis allowed identification of unknown early response genes (agnostic approach) and several teams identified gene signatures comprising dozens of genes for discriminating e.g. exposed from unexposed groups or different exposure heights [53, 55–59]. Other teams reduced the number of genes to the minimum required for dose estimation (below 10) [60–63]. With the invention of qRT-PCR for robust and sensitive quantitative GE analysis (gold-standard methodology), teams used only 1–4 genes [54, 64–73]. Large scale inter-laboratory comparison exercises initiated from NATO research task groups or the European biodosimetry network (RENEB) confirmed the applicability of this approach [74–77].

First published studies focused on GE changes at 24 h after irradiation, but this time window widened, covering 2 h to several days (supplemental table 1). Radiation-induced GE changes represent a biological process requiring a certain biological response time (several hours, supplemental table 1) which was recently recognized in the context of the RENEB 2019 exercise [76]. As that, it is a temperature sensitive process, and temperatures above or below 37 °C result in a several-fold decreased GE of widely used biomarkers for dose estimation which has to be considered in particular when performing field exercises [76].

Examinations on the association of GE with dose represents the topic of the majority of molecular radiobiological publications (see supplemental table 1, table 1). Dose rate dependent GE changes are less examined [78, 79] and only a few publications examined the impact of neutron (nuclear event [80–83]), or proton (mission to mars) exposures on mRNA GE in peripheral blood cells [84–86].

Radiation-induced GE saturates at a certain upper dose which seemed to be about 5 Gy after x-ray or γ-exposure (see supplemental tables 1, table 1). For highest mixed x-ray/neutron exposure at 3 Gy comparable DDB2 GE changes of about seven-fold to 3 Gy X–irradiation are reported [81]. These doses are sufficient enough for detection of severe haematological ARS (H-ARS) requiring hospitalization as discussed by Blakely et al in this issue. Several genes appear in particular sensitive for low radiation exposure and e.g. FDXR discriminated exposures of about 0.01–0.03 Gy (representative for a computer tomography) from unexposed individuals [87, 88].

GE measurements like all other biodosimetry assays require a reference to be interpreted. Knowing baseline expression and inter-individual variance of a certain gene, provides a measure discriminating unexposed from exposed individuals without need for individual pre-exposure controls, which has been shown by different groups already [55, 74, 89].

Suitable calibration curves are required in order to reflect exposure characteristics and time after exposure of a certain RN event. Strategies to overcome the exposure dependency and the time dependency are reflected in sections 2, 4.2 and 6 of this review, respectively.

International inter-assay studies allowed the comparison of [1] mean absolute difference of estimated doses relative to the true doses [2], number of dose estimates outside the 0.5 Gy interval as recommended for triage dosimetry as well as [3] the examination of merged doses into binary dose categories of clinical relevance and accuracy, sensitivity and specificity of the assays was compared. Regarding all three parameters, GE revealed a performance comparable to the micronucleus assay, but below the dicentric assay, still considered to represent the gold standard in biodosimetry [90]. However, GE measurements reveal early and high-throughput characteristics making them in particular attractive for RN scenarios involving hundreds or thousands of individuals. For instance, dose estimates can be provided from 1000 samples within 30 h [89] or 600 samples within 24 h [91].

Ex vivo whole blood experiments appear to reflect the in vivo situation regarding certain genes (e.g. FDXR, DDB2, WNT3 or POU2AF1), because irradiated T- and B-lymphocytes represent the origin of measured radiation-induced GE changes [92]. Other radiation-responsive genes examined in the peripheral blood after in vivo irradiation (e.g. CCR7, ARG2, CD177) are not seen ex vivo, indicating other radiation-responsive targets than the whole blood cells to be involved. Hence, using ex vivo experiments require in vivo models to estimate their significance.

Several groups continuously examined almost a dozen of genes over two decades. Details on this research are presented in supplemental table 1 and aggregated in table 1. Below, each of the genes will be presented following the biomarkers characteristics of section 3.

4.1.1. FDXR.

4.1.2. DDB2.

4.1.3. CDKN1A (P21).

4.1.4. GADD45.

4.1.5. BAX.

4.1.6. PCNA.

4.1.7. WNT3 and POU2AF1.

4.1.8. Gene signatures and combinations.

Examinations on the association of GE with health effects other than ARS are rather common outside the field of radiobiology and used for clinical purposes (see above) including e.g. associations with atherosclerosis [95], cancer [96, 97] and other clinical endpoints such as discrimination of sporadic from radiation-exposed thyroid or mammary cancer [98, 99] or identifying patients prone for developing late radiation toxicity [100, 101]. Prediction of radiation toxicity (e.g. normal tissue response, lung fibrosis) due to radiotherapy using GE changes is an approach examined in the context of radiation oncology [102–106]. Development of ARS even after whole body irradiation treatment of leukaemia patient is not observed, leaving ARS unobserved from the clinical point of view.

The radiobiological community has a strong interface to biology and physics, but less to the field for medical management of live threatening acute radiation health effects, which might contribute to the research focus on GE associations with radiation exposure (see above). GE biomarkers for radiation injury such as death/survival or ARS prediction, are less examined with some exceptions. Recently, two groups correlated GE with radiation-induced organ damage and survival in two large animal models [107, 108].

Discriminating ARS severity based on the clinical follow-up requires some medical expertise. Diagnostic guidance and treatment protocols for ARS have been established and updated [109]. The Medical Treatment Protocols for Radiation Accident Victims (METREPOL) document serves as a resource for physicians. Hematologic changes, such as the development of severe immune deficiency (due to lymphocytopenia and granulocytopenia) or bleeding due to thrombocytopenia, represent one of the challenges in appropriate clinical management of H-ARS. METREPOL categorizes H-ARS into five severity degrees based on BCC changes in the weeks that follow the exposure: no H-ARS (H0, referred to as baseline levels in METREPOL), low (H1), medium (H2), severe (H3) and fatal (H4) H-ARS. These severity degrees are associated with treatment decisions as outlined in METREPOL. Hence, predicting the H-ARS severity will ultimately provide important inputs for clinical diagnosis and treatment decisions. Since H-ARS develops with a delay over time, for prediction purposes H-ARS is supposed to be categorized based on the entire clinical follow-up of BCC spanning about 60 days. The alternative concept of correlating biomarkers with BCC changes to build H-ARS categories at the same early time points after irradiation bears the disadvantage to be not predictive and to just reflect the current clinical status which at early ARS stage (before the manifestation of the disease) renders this kind of research uninformative.

Furthermore, several authors use dose bands as a surrogate for the ARS severity, (e.g. ARS II = 2–4 Gy), thus, representing not a clinical, but an exposure related categorization [110, 111] with associated limitations (see section 2).

H-ARS severity degrees used herein, represent clinically relevant METREPOL based categories.

Using a baboon model combined with in vitro experiments and measurements on healthy donors and radiotherapy patients led to the prediction of the H-ARS with features as followed (for details see supplemental table 1 and table 1):

A combination of four genes was reported for early and high-throughput diagnosis of H-ARS severity [31]. This GE signature identifies [1] unexposed individuals (H-ARS 0) to conserve clinical resources for those requiring it [2], low-exposed individuals (H-ARS 1) requiring surveillance (late health effects) but no hospitalization or early treatment and [3] highly exposed individuals who will develop acute health effects, e.g. H-ARS 2–4 severity. The latter are in need of early intensive care. Merging existing H-ARS categories in those three categories appears appropriate to address urgent clinical questions such as prioritizing hospitalization as well as considering restricted medical resources. Pairwise redundant genes (FDXR/DDB2 and WNT3/POU2AF1) showing the same association with H-ARS severity degrees were employed in order to increase the robustness of the gene combination (e.g. difficulties with low baseline, figure 3).

Zoom In Zoom Out Reset image size

For identification of H-ARS 0 (unexposed individuals) all four radiation-induced GE changes relative to unexposed do not exceed a FC of 2. This is the FC considered to adjust for methodological variance.

For identification of H-ARS 1, a > two-fold up-regulation of FDXR and DDB2 and no GE changes of WNT3 and POU2AF1 below the methodological variance (FC 2 down-regulated) are expected.

For identification of H-ARS 2–4, a > two-fold up-regulation of FDXR and DDB2 and a pronounced down-regulation of WNT3 and POU2AF1 is expected. Based on currently available data, a 10-fold down-regulation of WNT3 or POU2AF1 predicts H-ARS 2–4 with a positive predictive value (PPV) around 100% [31, 112]. A > 2 and < 10-fold down-regulation of WNT3/POU2AF1 reduces the PPV from 100% to about 90% [112].

It is well known that radiation exposure causes changes in intracellular proteins and as a result of radiation-induced injury to tissues changes in blood and urine proteins [113]. In the case of radiation-induced intracellular proteins gamma-H2AX proteins have been advocated for use in biodosimetry [114]. Alternatively, Roy and colleagues proposed the use of organ-specific biomarkers as a biochemical approach to assess radiation dose and injury [115]. Table 2 illustrates a panel of organ-specific radiation biomarkers that have been proposed for use as biochemical dosimeters. Discovery and validation efforts typically involve initial evaluating blood samples from studies using rodent radiation models, followed by studies using blood/plasma samples derived from nonhuman primates (NHP) radiation studies and human radiation-therapy patients. In selected cases these candidate organ-specific proteomic biomarkers have been used for dose- and injury assessment in radiation accidents.

5.1.1. Gastrointestinal system biomarkers.

5.1.2. Hematopoietic system biomarkers.

5.1.3. Cutaneous system biomarkers.

5.1.4. DNA damage and repair pathway biomarkers.

5.1.5. Use of multiple proteomic biomarkers for dose assessment.

The ability to extend proteomic biomarkers findings from animal radiation studies to predict ARS severity using the METREPOL system [161], developed for use with humans, necessitates the establishment of a similar ARS severity scoring system in the animal models. Animal based ARS severity scoring systems have been now been established in rodent [148], Rhesus macaque [162] and baboon [39] model systems.

Early-phase blood plasma proteomic biomarkers have been established for the assessment of ARS severity in mouse [148] and baboon [40] model systems. In the baboon radiation model three time-windows were established for the H-ARS severity predictive algorithms: model 1 (0–2 days) includes CRP, IL‐13, and procalcitonin biomarkers; model 2 (2–7 days) includes CD27 (T and B cell surface biomarkers), Flt-3 ligand, SAA, and IL-6; and model 3 (7–28 days) includes CD27, SAA, EPO, and CD177 (neutrophil cell surface biomarkers) [40]. The use of proteomic-based ARS severity predictive algorithms can supplement clinical signs and symptoms, other biological dosimetry, and physical dosimetry endpoints to assess H-ARS risk severity.

Using data from an analysis of dosimetric and clinical data from a group of ARS patients exposed to γ- and neutron or γ-rays alone, Azizova and colleagues showed an application of the METREPOL-based severity scoring system to provide early-phase diagnostic assessment [163]. The constellation of early-phase prodomal clinical signs and symptoms, early changes in blood cell (i.e. neutrophils, lymphocytes) counts can provide ARS severity assessment. In 2016 Blakely and colleagues recommended that the METREPOL scoring system be supplemented using plasma proteomic biomarkers [164]; see updated figure 4.

Zoom In Zoom Out Reset image size

Many laboratories worldwide established their own gene set comprising 1–4 genes for dose estimation using qRT-PCR [54, 64–73]. Dose estimation requires a calibration curve comprising irradiated samples examined at the same time after irradiation as the potentially irradiated samples of some radiation victims. Due to the high automation degree, this extra burden can be handled sufficiently. Hence, dose estimates can be reported within hours after sample delivery as shown in several large-scale inter-laboratory comparison exercises [74–77]. Microarrays can be also used for this task. After selecting the blinded sample with the lowest GE value as the non-irradiated, GE values of a seven-gene as well as a two-gene combination and their internal calibration curves became re-adjusted and dose estimates on all blinded samples were performed [76]. Choosing the appropriate calibration curve (if available) in the absence of details regarding the radiation exposure reflects a difficulty of this approach. However, even in the absence of appropriate calibration curves, samples can be categorized based on their normalized GE and within the same time interval. Baseline values (from healthy donor samples) would identify unexposed individuals and from there, samples could be put into categories considering fold-changes (relative to unexposed) which, based on previous experience, reflect high radiation exposure demanding for hospitalization (acute severe health effects are expected) or lower radiation exposure without the need for hospitalization and early treatment, but requiring surveillance for late health effects (discussed in [31]).

As outlined in 4.1.8, hundreds or dozens of genes are employed for discriminating unexposed and exposed samples categorically (e.g. 0 vs 1.4 Gy) with signatures being specific for each examined time point. This approach demands a large variety of established gene sets to select the appropriate one. So far X- and γ-irradiation exposures are covered and other radiation qualities or dose-rate dependent examinations are largely missing.

Although challenging, one team introduced and evaluated a high throughput biodosimetry test system (REDI-Dx) continuously since 2007 [57, 77, 90, 91]. The signature consists of 15 radiation-responsive genes including e.g. FDXR, BAX, CDKN1A, PCNA, DDB2 and others [57, 91]). The test system provides dose estimates covering 0–10 Gy from 24 h to 7 days after irradiation [91]. Recently, authors examined the applicability of their tool for determining two dose categories of clinical significance: >2–6 Gy (hospitalization and intensive care like ICU and cytokine treatment required) and > 6 Gy (more aggravated H-ARS). ROC area over four different time points (e.g. 24 h and 72 h) and both dose categories (relative to unexposed healthy donors) always ranged between 0.94 and 1.0 [91]. Authors examined different confounders/parameter in 331 from altogether 656 unirradiated human subjects. Those were chronic conditions (e.g. allergy, asthma, diabetes, arthritis, pregnancy), skin burns (<10%, 10%–20%, >21% body surface), influenza, trauma (injury severity scale 10–14, 15–24 and >25), age (e.g. 18–21, 22–39, 40–64, >65), ethnicity and gender. Authors report negligible effects on identification of unexposed samples regarding these health conditions and demographic factors. Those examinations are missing for a total of 344 whole body irradiated samples which originated from NHP. Also, exclusion factors for patients comprised treatments with impact on whole BCC (and GE values) such as chemotherapy as well as cytokine inhibitor, -inducer or cytokine therapy (e.g. G-CSF) given prior to irradiation. Hence, dose estimates in patients with these treatments will probably fail.

Section 4.2 describes a combination of four genes for early and high-throughput diagnosis of H-ARS severity [31]. Figure 3 provides the onset of the time window which depending on the genes starts 2–8 h after irradiation. Over 3 days GE changes considerably, but even normal values are not observed three days after irradiation. Taking advantage of typically up- (FDXR and DDB2) or down-regulated genes (WNT3 and POU2AF1), different H-ARS severity degrees can be discriminated based on the fold-change and during the three days lasting time frame employing logistic regression analysis combined with ROC. Fold-differences in GE measured at different time points translate into a probability which is proportional to a PPV and NPV and a certain sensitivity and specificity [31]. Although validated in vitro, on healthy donors and on leukaemia patients, these data are based on 17 irradiated baboons and 200 unirradiated human healthy donor samples only [112]. Further validation of this approach on larger animal cohorts and more whole body irradiated leukaemia patients (with BCC changes similar to severe H-ARS) is required, but considering the underlying disease with strong impact on baseline GE [66].

Predicting H-ARS is an important step in diagnosis of the ARS. But other organ systems are also affected by ARS such as the gastrointestinal or the neurovascular system. Those are not covered by the diagnostic tool. However, the bone marrow is considered the most radiosensitive. Hence, identifying individuals who will develop H-ARS, automatically results in hospitalization of individuals who may also develop the ARS associated gastrointestinal or neurovascular syndrome or multi organ failure.

Several publications report on the impact of e.g. LPS (bacterial infection) on GE of certain (e.g. FDXR and CDKN1A), but not all genes (e.g. DDB2, GADD45, PCNA, see above). Also, demographic parameter such as ethnicity, age and gender could have an impact on GE and the associated allocation of unexposed or exposed individuals. Several publications report on a negligible effect of these factors for allocation of unexposed healthy donors (see section 6.2). This has been also examined on another 200 blood samples from 122 male and 88 female healthy donors and gender as well as age were found to lie well within the twofold differences in GE generally considered to represent control values [112]. Gene signatures are may be in particular robust across diverse disease states as shown by several groups [57, 91, 165]. Nevertheless, the association of single genes as outlined above with e.g. the TP53 signal transduction pathway makes them likely candidates being altered by other exposures or diseases as well. For instance, CDKN1A acts as a negative regulator of the cell cycle. Its expression is increased in response to various intra- and extracellular stimuli to arrest the cell cycle ensuring genomic stability as well as involved in diverse biological processes such as differentiation, cell migration, cytoskeletal dynamics, apoptosis, transcription, DNA repair, reprogramming of induced pluripotent stem cells, autophagy and the onset of senescence [166]. Furthermore, all treatments impacting the composition of peripheral BCC (see section 6.2) will change GE measurements. These thoughts highlight the requirement for considering GE changes context related. In a RN scenario, clinical parameter (e.g. vomiting, diarrhoea, erythema), physical measurements and further biological changes are expected (for details see Blakely et al in this issue). It is the context related view in its entirety, considering different aspects of the disease and not solely gene or protein expression changes, which finally results in the appropriate allocation of individuals to clinically relevant categories, we believe.

On-going efforts to advance the biochemical biodosimetry approach for radiation dose assessment has evaluated: (a) selection of biomarkers panel, (b) measurement of population baseline levels, and (c) consideration of potential confounders [121]. A similar approach is needed for a proteomic biomarkers panel applicable for assessment of ARS severity. In both of these cases consideration of special populations (i.e. children, immune comprised individuals, etc) need to be addressed. It is recommended that an evidence-based review of this approach be performed to obtain international consensus of this approach as the standard of diagnostics for early-phase assessment of potential life-threatening radiation exposures.

Use of peripheral blood as a convenient source for GE measurements, is widely accepted (see table 1 and supplemental table 1). Isolation of RNA from whole saliva samples as a non-invasive and easily accessible biofluid could represent an attractive alternative to blood. Saliva is a blood plasma ultra-filtrate and as that thought as a 'mirror of the body' [167, 168]. Although saliva has been already shown to contain RNA biomarkers for e.g. oral cancer diagnostic [169–171], it's use appears challenged because of an overwhelming bacterial contamination, associated with a low fraction of human RNA making a more complex workflow for meaningful RNA analysis necessary, which is currently in progress [172, 173]. Only a few studies report on radiation-induced changes in urine using e.g. mass spectrometry for metabolomic measurements in mice and NHP [174, 175]. Another study reports on the stability of miRNAs by collecting serial urine samples from patients with localized prostate cancer [176], indicating a developing field requiring further research.

In study scenarios where dozens or several hundred candidate genes and larger sample numbers are analysed, formats other than using 96 well plates have to be utilized for GE analysis. So-called customized low-density arrays (LDA, Thermo Fisher Scientific) or recent developments such as the 12k open array format (OA, Thermo Fisher Scientific) can be applied. In the LDAs, primer and probes for each gene are filled and lyophilized in each well, thus ensuring single qRT-PCR reactions and identification of up to 384 different genes per LDA. The 12k OA includes four metal slides with about 3000 holes each comprising primer and probes for each gene coated on the walls. LDA proved to be reliable, but 12k OA must be run in triplicate due to technical limitations inherent to this technology (company's recommendations and own experience). Besides these qRT-PCR based high-throughput methodologies, NGS and nanopore sequencing based technologies provide high-throughput capabilities due to multiplexing of hundreds of samples in one reaction combined with targeted sequencing of selected genes [65, 69, 88, 89].

Microfluidic technology and the concept of a 'lab on the chip' provides the promise to put the complex laboratory workflow from the blood drop to RNA-isolation, cDNA synthesis and qRT-PCR on a microfluidic card to be used on the point of care, e.g. in hospitals, where potentially radiation exposed individuals will arrive in a radiological or nuclear event. No established qRT-PCR based microfluidic cards exist so far, but several reports dealing with microfluidic mixing processes and work on a module for isolation of white blood cells from peripheral blood as a prerequisite for automation of GE assays on a microfluidic cartridge indicate some movement in this direction [177, 178]. Own experience with this technology reflect challenges in RNA isolation, limitations in the linearity dynamic range of GE values using a one-step or two-step qRT-PCR and miniaturization restrictions. After overcoming these challenges, a mobile platform will exist, however expenses per card have to be considered and multiplexing will be unlikely.

Recently, nanopore sequencing became introduced as a rapid (3 min for a total of 50 000 reads) and portable real-time biodosimetry platform, requiring some future improvements regarding sample processing and the bioinformatic pipeline for specific radiation-responsive transcript identification, to finally gain the full potential of this portable and rapid technology [69].

Lateral-flow immunoassay technology for protein detection is well established [179, 180]. Using this technology for nucleic acid detection is currently under development [181, 182].

Employing machine learning (artificial intelligence) for analysis of e.g. multi-omics biomarkers or medical imaging to improve clinical outcome predictions is one task [183–185]. However, there is ongoing activity using this technology for integration of multiple input data such as clinical signs and symptoms of ARS, radiation-induced cytogenetic and molecular data for accelerated and improved prediction of acute radiation health effects.

The proposed candidate proteomic biomarkers are all suitable for analysis using conventional enzyme-linked immunosorbent assay (ELISA)-based analysis. POC analysis using lateral-flow technology can support up to 3 targets from different organs as shown in tables 2 and 3. Multiplex analysis supporting up to 8–10 targets can be measured in a single well [146, 148]. Recent developments involving multiplex protein analysis using aptamer/SOMAmer reagents analysis or protein extension analysis show promise to extend these technologies to support biochemical biodosimetry analysis to assess both radiation dose and injury.