Gene signature of the post-Chernobyl papillary thyroid cancer

Although a rise in the incidence of thyroid cancer after the Chernobyl accident is evident [3, 22], the question of the potential molecular peculiarities of these induced tumours has not yet been resolved. Answering this question is not only of scientific interest, but also may expand our knowledge on how to manage internal radiation contamination.

In our study of post-Chernobyl PTC, we observed small but significant changes in the expression of 239 genes (p < 0.01) between tumours arising after exposure to low-dose radiation after the Chernobyl accident and sporadic PTCs. Our study is among the first to find differences in gene expression profiles between radiation-induced and sporadic PTC in patients matched for their ethnicity, place of residence, sex, histopathology, disease stage and age at diagnosis. Five previous transcriptomic studies comparing radiation-induced and sporadic thyroid cancer [16, 23‐26] were limited by the small number of studied patients, were not matched between sporadic and radiation-induced PTC due to differences in geographical distribution of the patients [23, 25, 26] and in PTC stage [25], and compared expression alterations in radiation-induced cancer with data repositories of sporadic PTC in adults [16]. The recently published study by Abend et al. [20] in which a well-characterized cohort of patients with radiation-induced PTC were analysed, showed radiation dose-dependent gene expression changes, but did not globally compare exposed and non-exposed patients. Our results support their general conclusion on the long-term differential gene expression in PTC arising after ionizing radiation exposure. This observation is also supported by recent results [15] demonstrating that PTC driver alterations are more prevalent in PTC in children who have been exposed to radiation.

Although to our knowledge our matched group of radiation-exposed patients and patients with sporadic PTC is optimal because of the availability of current biological samples, we are aware of a potential drawback in the ability to identify sporadic PTC developing in radiation-exposed patients. According to epidemiological estimation, about 29 % of patients in our ECR group may have developed PTC in the absence of radiation exposure [27]. The figure may possibly be even higher if the increased identification of PTC due to screening of the population is taken into account. We therefore cannot rule out the inclusion of some sporadic cancers in our ECR group. However, we were able to identify significant, although subtle, differences in gene expression profiles between ECR and non-ECR cancers. We can speculate that the inclusion of sporadic PTC may be one of the reasons for very subtle difference in gene expression with a fold change in the range 0.48 – 3.42. Also at the molecular level, in the ECR group we failed to separate tumours clustering closer to those in the non-ECR group either in the unsupervised multidimensional scaling principle component analysis (PCA) (Fig. 1) or in the more detailed supervised analysis. This leads us to speculate that the different gene expression between ECR and non-ECR tumours is rather related to radiation response than to carcinogenesis.

Our negative findings using PCA are in line with the results of Dom et al. [19], who in cooperation and in parallel with our group studied gene expression in normal thyroid tissue of radiation-exposed and non-exposed patients. They also were not able to show any differences using PCA, and only significance analysis of microarrays (SAM) with adjustment for age was able to identify 403 differentially expressed genes in normal thyroid tissue. Similarly in our study the difference between ECR and non-ECR tumours could only be detected after careful quality assurance, including gene filtering according to their expression level and variance. Thus, with such a stringent criterion, it is not surprising that there were only a few overlapping genes when we compared our 239 differentiating genes with the results of others. None of the top 15 candidate genes found to differ between radiation-induced and sporadic PTC by Port et al. [25] overlapped with ours or two other sets. No overlap was found either for the ten genes validated by us by qPCR. Only one gene (NEDD4L) identified by Detours et al. [23], four genes (ALDH6A1, TPD52L1, GPX1, ECE1) identified by Stein et al. [16] and two genes (MYO1C, IGF1R) identified by Ugolin et al. [26] were observed in our microarray gene signature.

Given that our multifactorial analysis of variance excluded the contribution of age differences and tumour pathology to the difference in gene expression profiles between the ECR group and non-ECR group, one can hypothesize that the genes identified here reflect a true difference between non-ECR and ECR PTC. However, our results also support an independent effect of a BRAF mutation on PTC gene expression profile. Interestingly, the effects of the presence of RET/PTC rearrangements were smaller [28]. This is consistent with the findings of previous studies showing differences between the effects of BRAF and RET/PTC alterations on gene expression in thyroid cancer [29, 30]. The frequency of RET/PTC rearrangements was not as high, and of BRAF mutation not as low as previously reported in post-Chernobyl PTC [7, 8, 12, 31]. This is also consistent with the fact that the median age of the patients at diagnosis was 17.7 years, which is distinctly higher than in previous post-Chernobyl cohorts [3], but similar to the age of the patients recently reported by Sassolas et al. [32]. The relationship between age at diagnosis and frequency of BRAF and RET/PTC alterations has also been previously identified in Ukrainian patients [33]. The requirement to age-match the patients with patients with sporadic PTC, which is more common in older children, in this study meant that patients in the ECR group were also slightly older than in the previous studies that did not use age-matched controls. In addition, 52 of our genes were validated by exon array analysis done in the partially independent and smaller set of tumours.

Environmental factors, such as differences in iodine deficiency, also need to be taken into consideration [34]. However in our study the place of residence of patients in the ECR and non-ECR groups were evenly distributed within different regions (oblast) and we consider that in a retrospective series cases this is the best available method to control for differences in iodine dietary status. Unlike other authors [18, 20], we did not show formal analysis of gene expression in relation to individual radiation doses provided by the CTB [35]. Although the Spearman’s dose–response correlation indicated a few significant genes (data not shown), due to uncertainty in radiation dose and possible inclusion of patients with sporadic PTC in the non-ECR group, we consider these data too biased. Furthermore, recently reported studies indicate more diverse gene expression profile with decreasing absorbed doses. This was observed in mouse thyroid cells after injection of different amounts of ²¹¹At or ¹³¹I radionuclides [36, 37]. It was hypothesized that at high absorbed doses, the DNA lesions might have been too complex to be properly repaired, resulting in reduced cellular response compared to that at lower absorbed doses.

An important feature of the investigated PTC patients was their young age, which contributed to the different PTC gene expression profile compared with adult patients (data not shown). However, our radiation gene signature contained both genes, which did and did not contribute to the tumour/normal difference in the studied patients (data not shown) [38, 39]. Thus, our study defined the difference in gene expression related to radiation exposure, and the functional consequences of this need to be defined. To understand the underlying biological mechanisms, the genes confirmed by qPCR need to be examined in independent PTC cases in relation to G2/M cell cycle arrest. The simultaneous lower expression of CDK1 and RAD51AP may represent impaired repair of the radiation-induced DNA damage in ECR patients. The expression of CDK1 in fibroblasts is reduced in response to radiation [40], and its suppression is essential for DNA damage-induced G2 arrest [41]. CDK1 is required for efficient 5′ to 3′ resection of double-strand break ends, and for the recruitment of the single-stranded DNA-binding complex, RPA and the RAD51 recombination protein [42]. Decreased RAD51AP, encoding an enhancer of RAD51, observed in tumours from ECR patients is consistent with this suggestion, as genetic ablation of RAD51AP1 leads to enhanced sensitivity to chromosome aberrations upon DNA damage [43]. RAD51AP1-depleted cells have deficits in recombination-based repair of a DNA double-strand break, and exhibit chromatin breaks both spontaneously and upon DNA-damaging treatment [44].

The simultaneous increase in expression of HDAC11 in ECR-related PTC creates a link to transcriptional repression and epigenetic landscaping [45], and can be interpreted as concordant with both CDK1 and RAD51AP1 decreases as the latter is regulated by E2F family of transcription factors, while histone deacetylases interact with RB-E2F to inhibit gene transcription and are activated by radiation [46]. This effect may be stronger at the basic higher gene expression level. The reduced expression of PPME1 may also be related to the repair of gamma irradiation-induced DNA damage, which is regulated not only by PP1, but also by PP2A phosphatase inhibition [47]. Its protein product, protein phosphatase methylesterase 1, is regarded as a key molecule that sustains the activation of ERK activity in cancer cells via inhibition of PP2A [47, 48]. The higher expression of this gene group in thyroid cancers of the ECR group may lead to the higher activation of MAP cascade downstream of growth factors, but upstream of RAF and facilitate neoplastic transformation towards PTC [10]. Indeed, 4 of 14 genes known to modulate PP2A were significantly changed in ECR-related tumours (Supplementary Table S2). These effects may have been further enhanced be upregulation of ERBB2 and THRA (thyroid hormone receptor A) in the ECR group. Recently THRA-rs939348 was confirmed as a risk factor for DTC [49], and one may speculate that its increased expression in ECR tumours is a persistent response to radiation DNA damage which may cooperate with other genes in DTC development.

Obviously, a number of other potential speculative explanations for the observed gene expression differences could be presented. It cannot be excluded that cancer induced by a single dose of radiation shows a difference in cellular homogeneity (increased number of multiplied transformed cells and their desynchronization), kinetics of progression, or even in the tumour size at diagnosis.

An important study of the molecular biology of thyroid cancer discussing the results of The Cancer Genome Atlas (TCGA) has recently been published [28]. The results of the study indicate the relatively low number of novel genomic events in PTC compared to the previous knowledge and indicate the presence of subtypes, mainly related to the type of initiating somatic abnormality. It is an obvious next step to apply genomic sequencing to analyse in-depth the association of these subtypes and heterogeneity related to different initiating mutations with the profile of radiation-induced PTC. It is important to note that the expression of all genes characteristic of ECR PTCs according to our signature in the PTCs investigated by TCGA was high.

The important question arises as to whether subtle differences between the profiles of radiation-induced and sporadic PTCs have any clinical significance. Probably they do not reflect profound differences in the underlying disease, but rather different disease kinetics, cellular composition or – most interestingly – additional molecular mechanisms operating in the radiation-induced cancer. The proposed classifier is not sufficient in itself to distinguish the cancers induced by low-dose radiation from sporadic cancers, and our results indicate that the effect of radiation is similar in scale to many other factors influencing the variability of gene expression in PTC. We did not find any gene expression differences profound enough to influence the clinical course of the disease, and this is in line with the clinical observations indicating similar prognosis in post-Chernobyl childhood PTC [3, 50]. However, we interpret the differences observed by us as an excellent starting point to assess the importance of genes constituting the radiation signature in the pathogenesis of PTC.

In conclusion, we report significant, but subtle, differences in gene expression in the post-Chernobyl PTC that are associated with low-dose radiation exposure. Since the population exposed to low-dose thyroid radiation (either medical or accidental) is increasing, the study may serve as a basis for further studies on the susceptibility of the thyroid to low-dose radiation.