Results

Ex vivo irradiation of human peripheral blood provides an initial signature capable of classifying samples by dose.

Blood from ten unrelated donors was split into aliquots and exposed to 0, 0.5, 2, 5, or 8 Gy radiation, and then incubated for either 6 or 24 hours before RNA was extracted, labeled, and hybridized to whole genome microarrays. A nearest-centroid classifier comprised of 74 array features was identified that was capable of correctly classifying 98% of samples as being exposed to 0, 0.5, 2 or ≥5 Gy.

The ex vivo signature is robust in new data sets and shows minimal confounding by sex or smoking.

In this study, we examined the effect of cigarette smoking on gene expression in the peripheral blood, and its potential effect on radiation response and classification signatures. We used a single 6-hour assay time, as gene expression and predictive power of the gene expression signatures was not found to be significantly different between 6 and 24 hours post-exposure in or previous studies above. This study also used a new low dose of 0.1 Gy, as we reasoned that if smoking resulted in constitutive alteration of some DNA damage signaling pathways, confounding might be more likely among controls and radiation exposures at the low end of the dose range.

We recruited new volunteers for this study as either non-smokers or smokers (one pack a day or more). We analyzed results from 6 male smokers, 6 male non-smokers, 6 female smokers, and 6 female non-smokers. There were broad differences in the expression of a large number of genes when we compared smokers and non-smokers (Top figure). No radiation dose-response relationship was obvious among the hierarchically clustered smoking-associated genes. In contrast, the “consensus” set of 74 dose-responsive genes previously defined from our ex vivo work were clearly dose responsive in the smokers and did not show any obvious differences in response between smokers and non-smokers or males and females (Bottom figure).

Importantly, prediction of exposure dose using the 74-gene set was not affected by smoking status and extended to prediction of the newly introduced 0.1 Gy exposed samples (Table 1), demonstrating the robustness of this signature. Both linear discriminant analysis and 3-nearest neighbors predicted the exposure dose of 99% of the new samples correctly.

Table 1

Dose (Gy)

Sensitivity

Specificity

0

1

1

0.1

1

1

0.5

1

0.99

2

0.96

1

For more details and analyses, see the full study here.

Ex vivo gene expression signature predicts dose to patients irradiated in vivo.

To extend our studies to in vivo human exposures, we obtained blood samples from patients undergoing total body irradiation (TBI) at Memorial Sloan Kettering Cancer Center (MSKCC) for a variety of diseases including AML, ALL, CML, DLBL, MCL, multiple myeloma, and T-cell ALL. Blood was collected several hours before the initial radiation treatment, then at 4 hours after the first 1.25-Gy fraction and at 24 hours after the first fraction, and the RNA was subjected to whole genome microarray analysis. The 24-hour samples included exposure to three 1.25 Gy fractions. To further test for potential differences between the TBI patients and a healthy population, we included 14 samples drawn from healthy donors and processed identically to the TBI samples.

We found a strong overlap between in vivo and ex vivo responses. When we applied the74-gene signature derived from our earlier ex vivo work to the in vivo TBI samples (and healthy controls), it predicted the dose correctly in 96-99% of the samples depending on the algorithm applied. As an example, the performance of the nearest centroid classifier is detailed in the table below, where it can be seen that one 1.25 Gy sample was misclassified as having received 3.75 Gy, while all unexposed controls (healthy or TBI patient) were correctly classified.

Table 2

Dose (Gy)

Sensitivity

Specificity

0

1

1

1.25

0.95

1

3.75

1

0.98

Visualization of expression of the 74-gene signature in ex vivo and in vivo samples by multi-dimensional scaling (MDS) further illustrates that the disease status of TBI patients does not affect the expression of the genes in our consensus signatures. The expression of these genes in controls depends on whether or not the samples were cultured ex vivo, and not on the health status of the donors. The lateral shift between in vivo and ex vivo samples is similar to that seen previously with different times in culture. The 2 Gy ex vivo dose is also shown in the figure to the right for a comparison of the radiation effect.

Overall the results of this study strongly suggest that 1) the response in blood cells of healthy donors is not intrinsically different from that in individuals who have had cancer, making cancer patients a reasonable model for in vivo human exposures, and 2) ex vivo irradiation of blood is a useful model for predicting the radiation response in humans in vivo. For more details and analyses, see the full study here.

Effects of internal emitters - 137Cesium

Mice were injected with soluble 137Cs at Lovelace Respiratory Research Institute to provide a total body exposure to gamma and beta radiation. Five RNA samples per point from animals sacrificed at 2, 3, 5, 20, and 30 days after injection were hybridized to microarrays along with RNA from control animals. The analysis identified a large number of genes with differential expression (p<0.001, false discovery rate (FDR) <1%) compared to controls at each of the sacrifice times. These times corresponded to an average dose to the animals of approximately 2 Gy (2 days), 2.7 Gy (3 days), 4 Gy (5 days), 9.5 Gy (20 days), and 9.9 Gy (30 days). Only 5 genes were significantly differentially expressed at all five times tested. At later times / higher doses there was a marked trend toward a larger number of genes responding and an increasing proportion of down-regulated genes.

Gene ontology analysis of the data suggested that the biological processes that were over-represented among the responding genes changed over time, as both the total accumulated dose increased and the dose rate of exposure decreased. The first functions affected were largely related to histones and chromatin-related functions, cytoskeletal functions, hemostasis and regulation of body fluid levels. These functions were no longer significantly over-represented among differentially expressed genes after 20 days, however. The functions that appeared later in the experiment were largely related to mitochondrial and ribosomal structure and function, differentiation, proliferation, and cell death. Despite the dynamic nature of the cellular processes that appeared to respond to internal 137Cs, some functions remained significantly over-represented among responding genes (although not necessarily by all the same genes) at all times measured. These functions were mainly related to immune response and leukocyte-specific functions. For a more detailed report, see the full study here. Ongoing experiments will investigate the effects of different amounts of 137Cs initially deposited in the animals.

Low dose-rate studies

We have used the modified low dose rate irradiator to expose blood from healthy human donors to either acute (1 Gy/min) or low dose rate (LDR, 3 mGy/min) x rays. We measured whole-genome gene expression in RNA extracted from these samples 24 hours after the start of exposure and found several different dose-response patterns. Most genes were overexpressed by both exposure rates, but with less response to LDR (C in the Figure). Some genes were similarly under expressed after acute and to a lesser extent LDR exposure (Figure, A). A small set of genes showed responses predominantly to LDR exposure (Figure, B and D). Our previously derived dosimetric signatures were able to predict the dose to these samples regardless of dose rate, classifying 97% of the samples correctly. Other genes could distinguish between acute and LDR exposures, however. We are working with the Informatics Core to develop dose rate classifiers, as the rate of exposure will modify the extent of a radiological injury, making this important information for triage. Similar response patterns were found in the blood of mice exposed to the same dose and dose-rate combinations. Details of these experiments with human blood and mice have been published.

Neutron responses

Understanding the impact of the neutron component of radiation exposure on biodosimetry endpoints would be crucial in the event of an improvised nuclear device detonation. In initial studies comparing the gene expression response to IND-spectrum neutrons with the response to photons, we focused on general and gene-specific Relative Biological Effectiveness (RBE). We irradiated human blood samples with 0.1, 0.3, 0.5, and 1 Gy of either IND-spectrum neutrons or photons. At 1 Gy, we found an average RBE of 1.5 for gene expression, with considerable variation from gene to gene, suggesting a differential relative effect of neutrons on different aspects of the cellular response. We identified 325 genes that were differentially expressed (p<0.001; FDR ≤ 3.8%) as a function of x-ray dose, and 224 genes differentially expressed (p<0.001 FDR ≤ 5.4%) as a function of neutron dose. 125 genes were common to both the x-ray and neutron dose-responsive sets, and had an average RBE of 1.8 across all genes and doses, with the majority of genes showing a neutron RBE >1.

When we examined RBE as a function of dose, we found that for most genes, the RBE relationship was fairly constant across this dose range, with some genes showing a slight increase in RBE with increasing dose. Interestingly, a small set of genes did not appear to show an effect of LET, and had an RBE close to 1 at all doses tested. Such differential responses could help to distinguish a high LET component in a mixed field exposure, which would impact on the extent of injury and the selection of appropriate treatment. A manuscript describing this work is currently in preparation.