Ionising radiation is a proven human carcinogen. Most studies have analysed the effects of high dose radiation such as atomic bomb survivors in Japan, people exposed during the Chernobyl nuclear accident, patients undergoing radiation therapy, uranium miners, etc. However, it has been difficult to measure and assess the risk of cancer in people exposed to lower doses of ionising radiation such as the people living at high altitudes, who are exposed to more natural background radiation from cosmic rays than people at sea level. In our study, one of the points of interest was that as time progressed and even though the biological processes were very different across doses at 3 h, the response - by the set of common DE genes- to both 10 cGy (low) and 100 cGy (high) doses of X-ray radiation was similar in terms of the biological pathways (mostly cell cycle processes) affected at 24 h (Figure 11). Most of the processes at 24 h under both doses are associated with mitosis checkpoints. We have shown that the set of common DE genes account for a large percentage of the DE genes at each comparison. Therefore we claim that even 10 cGy radiation is not ‘low’ when analysed from a biological processes viewpoint. It is possible that cells exposed to the LDIR have better survival compared to the cells exposed to the HDIR after 24 h, despite the similarity in the processes at 24 h. It would be worth investigating this idea. It must be noted that our results may be specific to the model used in our study, but comparison to other LDIR studies have shown similar results in terms of the biological processes at each time and dose [29, 30, 43].
In this analysis we applied systems biology network analysis methods to study the difference in the effects of two doses of radiation on skin cells at different time points. This computational method of analysis helps to identify transcriptional differences between two very similar conditions. Our method of analysis is a way of fine tuning down to the sets of genes that may be key players at a certain time and, hence, by investigating the biological pathways of those key genes, we were able to identify the dominating processes at each time point. So it is the difference between coarse tuning - using DE gene selection or clustering and then identifying biological pathways technique - and fine tuning - our method of analysis via topological analyses.
Eukaryotic cells have many biological mechanisms to identify and repair damaged DNA to preserve genomic integrity. These mechanisms include the activation checkpoints and induction of cell cycle arrest, to allow the cell time to repair the damage. Cell cycle arrest can be triggered at G1/S, intra-S and G2/M phases. We saw these processes being dominant at various dose and time points in Figure 11. Microarray expression analysis of skin cells under normal conditions would probably show similar biological processes over time as cell cycle activity in skin cells is very high. There aren’t any reports of microarray expression of normal skin cells over time, to the best of our knowledge, to verify this idea. It is due to topological analyses that we were able to extract genes which have been reported to be highly associated with cancer, such as BRCA1, ageing, immune response, etc., which would not have been present under normal conditions. Methods like ours can help in generating testable hypotheses while dealing with high throughput data.
There has been a report by Voy et. al that has used a systems biology approach for analysing the effects of low dose ionising (10 cGy) radiation . Voy et. al identify sets of co-expressed genes by identifying ‘cliques’ in co-expression networks. To the best of our knowledge, Voy et. al is the only group, other than ours, that has analysed LDIR from a networks perspective. Not only is our experimental design very different but our computational analysis is also different. Our network topological analyses seek to identify the differences between two radiation doses networks, i.e. identify the differences across very similar conditions. We consider 10 cGy and 100 cGy ‘similar’ in the sense that both are ionising xradiation and the conditions are similar although the magnitudes are different. Furthermore, we use only local network properties to distinguish between the doses. The reason for this is that global network characteristics of very similar conditions (represented as networks) do not have enough power to distinguish between conditions. For example, consider two oranges - one sweet and one sour. Both look the same in terms of structure and texture, however, they differ in taste. If we built a network to represent the sweet orange and the sour orange, the global network characteristics, such as network density, mean network connectivity, etc. would be similar between the two orange networks. However, since there is a difference in taste, there will be some differences between these two networks which would be local network properties. Therefore, we do not use global properties to identify the differences between dose 10 cGy and 100 cGy in this manuscript. Other reports have also shown how global properties cannot distinguish between very similar networks [7, 44].
Systems biology methods like ours are in high demand as the differences between many conditions, be they neurodegenerative diseases, brain diseases, different kinds of cancers, different degrees of disease severity, etc., are very subtle and cannot be easily highlighted using the usual off-the-shelf clustering or biological pathways identification algorithms. Comparing networks of any kind - social, biological, etc. - is a difficult and ongoing field of research. Comparisons of gene coexpression networks by identifying hub genes or by looking at the neighbourhoods of gene in networks that share few common genes and have many different genes between them (both are cases where there are very obvious differences between networks that can be detected using simple methods), has been previously reported. However, comparing different networks representing very similar conditions and trying to identify the differences between them is a difficult task. We have presented the application of a method that can aid in this task.