Machine learning methodology for high throughput personalized neutron dose reconstruction in mixed neutron + photon exposures

Validation of a High-Throughput Dicentric Chromosome Assay Using Complex Radiation Exposures

Validation of biodosimetry assays is routinely performed using primarily orthovoltage irradiators at a conventional dose rate of approximately 1 Gy/min. However, incidental/ accidental exposures caused by nuclear weapons can be more complex. The aim of this work was to simulate the DNA damage effects mimicking those caused by the detonation of a several kilotons improvised nuclear device (IND). For this, we modeled complex exposures to: 1. a mixed (photons + IND-neutrons) field and 2. different dose rates that may come from the blast, nuclear fallout, or ground deposition of radionuclides (ground shine). Additionally, we assessed whether myeloid cytokines affect the precision of radiation dose estimation by modulating the frequency of dicentric chromosomes. To mimic different exposure scenarios, several irradiation systems were used. In a mixed field study, human blood samples were exposed to a photon field enriched with neutrons (ranging from 10% to 37%) from a source that mimics Hiroshima's A-bomb's energy spectrum (0.2-9 MeV). Using statistical analysis, we assessed whether photons and neutrons act in an additive or synergistic way to form dicentrics. For the dose rates study, human blood was exposed to photons or electrons at dose rates ranging from low (where the dose was spread over 32 h) to extremely high (where the dose was delivered in a fraction of a microsecond). Potential effects of cytokine treatment on biodosimetry dose predictions were analyzed in irradiated blood subjected to Neupogen or Neulasta for 24 or 48 h at the concentration recommended to forestall manifestation of an acute radiation syndrome in bomb survivors. All measurements were performed using a robotic station, the Rapid Automated Biodosimetry Tool II, programmed to culture lymphocytes and score dicentrics in multiwell plates (the RABiT-II DCA). In agreement with classical concepts of radiation biology, the RABiT-II DCA calibration curves suggested that the frequency of dicentrics depends on the type of radiation and is modulated by changes in the dose rate. The resulting dose-response curves suggested an intermediate dicentric yields and additive effects of photons and IND-neutrons in the mixed field. At ultra-high dose rate (600 Gy/s), affected lymphocytes exhibited significantly fewer dicentrics (P < 0.004, t test). In contrast, we did not find the dose-response modification effects of radiomitigators on the yields of dicentrics (Bonferroni corrected P > 0.006, ANOVA test). This result suggests no bias in the dose predictions should be expected after emergency cytokine treatment initiated up to 48 h prior to blood collection for dicentric analysis.

A machine learning method for improving the accuracy of radiation biodosimetry by combining data from the dicentric chromosomes and micronucleus assays

A large-scale malicious or accidental radiological event can expose vast numbers of people to ionizing radiation. The dicentric chromosome (DCA) and cytokinesis-block micronucleus (CBMN) assays are well-established biodosimetry methods for estimating individual absorbed doses after radiation exposure. Here we used machine learning (ML) to test the hypothesis that combining automated DCA and CBMN assays will improve dose reconstruction accuracy, compared with using either cytogenetic assay alone. We analyzed 1349 blood sample aliquots from 155 donors of different ages (3-69 years) and sexes (49.1% males), ex vivo irradiated with 0-8 Gy at dose rates from 0.08 Gy/day to ≥ 600 Gy/s. We compared the performances of several state-of-the-art ensemble ML methods and found that random forest generated the best results, with R² for actual vs. reconstructed doses on a testing data subset = 0.845, and mean absolute error = 0.628 Gy. The most important predictor variables were CBMN and DCA frequencies, and age. Removing CBMN or DCA data from the model significantly increased squared errors on testing data (p-values 3.4 × 10^-8 and 1.1 × 10^-6, respectively). These findings demonstrate the promising potential of combining CBMN and DCA assay data to reconstruct radiation doses in realistic scenarios of heterogeneous populations exposed to a mass-casualty radiological event.

A MATLAB-based program for three-dimensional quantitative analysis of micronuclei reveals that neuroinflammation induces micronuclei formation in the brain

The micronucleus is known to be a biomarker for genomic instability, which is a hallmark of tumors and aging. Normally, micronuclei are produced by segregation errors and mechanical stresses arising from dividing or migrating cells, leading to activation of the innate immune response pathway. Although micronuclei often emerge in damaged tissues, the quantitative procedure for analyzing micronuclei accurately has been problematic. Here, we introduce a novel MATLAB-based program for quantifying micronuclei (CAMDi: calculating automatic micronuclei distinction) in vitro and in vivo. CAMDi is adaptable to various experimental imaging techniques and is useful for obtaining reproducible data. CAMDi enables us to measure the accurate size of micronuclei from the three-dimensional images. Using CAMDi, we revealed a novel link between the emergence of micronuclei and neuroinflammation. We found that inflammatory stimulation does not increase the number of micronuclei in primary neurons. On the other hand, the administration of lipopolysaccharide into mice slightly increases micronuclei formation in neurons of the hippocampus region. These findings demonstrate that neuronal micronuclei formations are induced by an inflammatory response in a non-cell-autonomous manner. We provide a novel tool, CAMDi, to quantify micronuclei and demonstrate that neuronal micronuclei are produced not only by the cell-autonomous process but also by the intercellular communication associated with neuroinflammation in vivo.

Cytogenetic Damage of Human Lymphocytes in Humanized Mice Exposed to Neutrons and X Rays 24 h After Exposure

Detonation of an improvised nuclear device highlights the need to understand the risk of mixed radiation exposure as prompt radiation exposure could produce significant neutron and gamma exposures. Although the neutron component may be a relatively small percentage of the total absorbed dose, the large relative biological effectiveness (RBE) can induce larger biological DNA damage and cell killing. The objective of this study was to use a hematopoietically humanized mouse model to measure chromosomal DNA damage in human lymphocytes 24 h after in vivo exposure to neutrons (0.3 Gy) and X rays (1 Gy). The human dicentric and cytokinesis-block micronucleus assays were performed to measure chromosomal aberrations in human lymphocytes in vivo from the blood and spleen, respectively. The mBAND assay based on fluorescent in situ hybridization labeling was used to detect neutron-induced chromosome 1 inversions in the blood lymphocytes of the neutron-irradiated mice. Cytogenetics endpoints, dicentrics and micronuclei showed that there was no significant difference in yields between the 2 irradiation types at the doses tested, indicating that neutron-induced chromosomal DNA damage in vivo was more biologically effective (RBE ∼3.3) compared to X rays. The mBAND assay, which is considered a specific biomarker of high-LET neutron exposure, confirmed the presence of clustered DNA damage in the neutron-irradiated mice but not in the X-irradiated mice, 24 h after exposure.