Cross-species conserved miRNA as biomarker of radiation injury over a wide dose range using nonhuman primate model

Abstract

Multiple accidents in nuclear power plants and the growing concerns about the misuse of radiation exposure in warfare have called for the rapid determination of absorbed radiation doses (RDs). The latest findings about circulating microRNA (miRNAs) using several animal models revealed considerable promises, although translating this knowledge to clinics remains a major challenge. To address this issue, we randomly divided 36 nonhuman primates (NHPs) into six groups and exposed these groups to six different radiation doses ranging from 6.0-8.5 Gy in increments of 0.5 Gy. Serum samples were collected pre-irradiation as well as three post-irradiation timepoints, namely 1, 2 and 6 days post-total body irradiation (TBI). Generated from a deep sequencing platform, the miRNA reads were multi-variate analyzed to find the differentially expressed putative biomarkers that were linked to RDs, time since irradiation (TSI) and sex. To increase these biomarkers' translational potential, we aligned the NHP-miRNAs' sequences and their functional responses to humans following an in-silico routine. Those miRNAs, which were sequentially and functionally conserved between NHPs and humans, were down selected for further analysis. A linear regression model identified miRNA markers that were consistently regulated with increasing RD but independent TSI. Likewise, a set of potential TSI-markers were identified that consistently shifted with increasing TSI, but independent of RD. Additional molecular analysis found a considerable gender bias in the low-ranges of doses when the risk to radiation-induced fatality was low. Bionetworks linked to cell quantity and cell invasion were significantly altered between the survivors and decedents. Using these biomarkers, an assay could be developed to retrospectively determine the RD and TSI with high translational potential. Ultimately, this knowledge can lead to precise and personalized medicine.

Fig 1. Study design.

The flow diagram shows the longitudinal plan for collecting blood samples for circulating miRNA pre- and post-irradiation from NHPs. The table displays the sample size (M/F) and the survival outcome.

Fig 2. A flow chart to depict the process to deliver miRNAs that were not only sequential homologues between humans and NHPs, but were also likely to respond similarly to irradiation across these two species.

Fig 3. A flow diagram to explain the algorithm named 2BDP (Biomarker Discovery Process at Binomial Decision Point) that we used to find the best fitting model to predict RRiF.

Fig 4. Putative miRNA markers of irradiation.

All miRNAs sequentially and functionally conserved between humans and NHPs. (A) Time since irradiation (TSI)-independent, radiation dose (RD)-miRNA markers. Here, the x-axis and y-axis represented the radiation dose and normalized log2 (Fold change), respectively. The fold changes were computed on the pre-irradiation baseline controls. All these miRNAs displayed a consistent shift (p < 0.05 in linear regression curve) across dosimetry independent of TSI. * p < 0.05; *** p < 0.001. (B) RD-independent, TSI-miRNA markers. Here, the x-axis and y-axis represented the TSI and normalized log2 (Fold change), respectively. The fold changes were computed on the pre-irradiation baseline controls. All these miRNAs displayed consistent shift (p < 0.05 in linear regression curve) across TSI. * p<0.05; ** p<0.05; *** p<0.001.

Fig 5. Functional association among the differentially expressed miRNA cluster linked to RD and TSI.

The color keys of log2 (Fold change) were added at top left corners. In the networks, the oval and rectangular shaped nodes represent miRNA and biofunctions, respectively. The edges represent the relationship between two nodes: the solid lines represent their association, and pointed arrowheads denote the activating relationship between the two nodes. Of these, miRNAs that appeared most connected to the neighbors were annotated in the red colored stars for > 90% percentile, and green colored stars for > 75% percentile. (A) Hierarchical cluster of 17 miRNAs that were differentially expressed by the cumulative effects of RD and TSI. These miRNAs are sex-independent markers. (B) Network cluster of the same 17 miRNAs that were differentially expressed by RD*TSI. (C) Hierarchical cluster of 11 miRNAs that were differentially expressed by the cumulative effects of Sex, RD and TSI. (D) Network cluster of the same 11 miRNAs that were differentially expressed by Sex*RD*TSI.

Fig 6. Functional association among the differentially expressed miRNA cluster linked to RRiF.

The color keys of log2 (Fold change) were added at the top left corners. In these networks, the oval and rectangular shaped nodes represented miRNA and biofunctions, respectively. The edges represented the relationship between two nodes, the solid lines represented their association, and pointed arrowheads denoted the activating relationship between the two nodes. Red colored stars annotate miRNAs that were most connected to the neighbors for > 90% percentile, and green colored stars for > 75% percentile. (A) Hierarchical cluster of 21 differentially expressed miRNAs linked to RRiF. (B) Network cluster of the same 21 differentially expressed miRNAs linked to RRiF. (C) Hierarchical cluster of 28 differentially expressed miRNAs linked to the cumulative effects of radiation dose and RRiF. (D) Network cluster of the same 28 differentially expressed miRNAs linked to RD*RRiF.

Fig 7. An miRNA panel to predict RRiF.

Two putative top performing markers were listed herein. (A) This three-miRNA enriched panel was determined by k-fold routine of the 2BDP algorithm. AUC ROC plot is included herewith. (B) A seven-miRNA enriched panel was determined by RSBMR routine of the 2BDP algorithm. AUC ROC plot is included herewith.