CLIP2 As Radiation Biomarker in Papillary Thyroid Carcinoma

ORIGINAL ARTICLE CLIP2 as radiation biomarker in papillary thyroid carcinoma

M Selmansberger1, A Feuchtinger2, L Zurnadzhy3, A Michna1, JC Kaiser4, M Abend5, A Brenner6, T Bogdanova3, A Walch2, K Unger1,7, H Zitzelsberger1,7 and J Hess1,7

A substantial increase in papillary thyroid carcinoma (PTC) among children exposed to the radioiodine fallout has been one of the main consequences of the Chernobyl reactor accident. Recently, the investigation of PTCs from a cohort of young patients exposed to the post-Chernobyl radioiodine fallout at very young age and a matched nonexposed control group revealed a radiation-specific DNA copy number gain on chromosomal band 7q11.23 and the radiation-associated mRNA overexpression of CLIP2. In this study, we investigated the potential role of CLIP2 as a radiation marker to be used for the individual classification of PTCs into CLIP2- positive and -negative cases—a prerequisite for the integration of CLIP2 into epidemiological modelling of the risk of radiation- induced PTC. We were able to validate the radiation-associated CLIP2 overexpression at the protein level by immunohistochemistry (IHC) followed by relative quantification using digital image analysis software (P = 0.0149). Furthermore, we developed a standardized workflow for the determination of CLIP2-positive and -negative cases that combines visual CLIP2 IHC scoring and CLIP2 genomic copy number status. In addition to the discovery cohort (n = 33), two independent validation cohorts of PTCs (n = 115) were investigated. High sensitivity and specificity rates for all three investigated cohorts were obtained, demonstrating robustness of the developed workflow. To analyse the function of CLIP2 in radiation-associated PTC, the CLIP2 gene regulatory network was reconstructed using global mRNA expression data from PTC patient samples. The genes comprising the first neighbourhood of CLIP2 (BAG2, CHST3, KIF3C, NEURL1, PPIL3 and RGS4) suggest the involvement of CLIP2 in the fundamental carcinogenic processes including apoptosis, mitogen-activated protein kinase signalling and genomic instability. In our study, we successfully developed and independently validated a workflow for the typing of PTC clinical samples into CLIP2-positive and CLIP2-negative and provided first insights into the CLIP2 interactome in the context of radiation-associated PTC.

Oncogene (2015) 34, 3917–3925; doi:10.1038/onc.2014.311; published online 6 October 2014

INTRODUCTION alterations in PTC are point mutations of the BRAF gene (V600E) One of the major consequences of the Chernobyl nuclear accident and various variants of RET gene rearrangements, all of which lead in 1986 has been a significant increase in the incidence of to a constitutive activation of the mitogen-activated protein 8 papillary thyroid carcinomas (PTCs) among children exposed to kinase (MAPK) pathway. The frequency of RET/PTC rearrange- the radioiodine fallout, particularly to iodine-131.1 To date, PTC ments (that is, RET/PTC1 and RET/PTC3) was associated with 4 radiation exposure levels in a study of the atomic bomb survivors has developed in 4000 individuals who were children or 9–11 adolescents at the time of exposure.2 Thus, young age at exposure and some but not all studies of post-Chernobyl PTC. However, is a significant risk factor for the development of radiation- RET/PTC3 rearrangements have also been observed with similar frequencies in sporadic PTCs from young patients, indicating a induced PTC.3 In order to delineate radiation-associated effects, it relation with young age of PTC onset.11,12 A more recent is crucial that the tumour cohorts of exposed and nonexposed 13 4 publication by Ricarte-Filho et al. reported a higher frequency cases are matched on age and other factors as closely as possible. of fusion oncogenes in radiation-induced PTCs, including rare TRK This approach was enabled by the Chernobyl Tissue Bank and BRAF rearrangements and the recently discovered ETV6– (CTB, www.chernobyltissuebank.com) that systematically collects NTRK3 kinase fusion oncogene. These promising findings require thyroid tumour tissue samples from residents who lived in the further validation in independent cohorts. We recently reported a contaminated regions of Ukraine and the Russian Federation at radiation-specific DNA copy number gain on the chromosomal the time of the accident. The CTB collection also includes a band 7q11.23 and a radiation-associated mRNA overexpression of substantial number of thyroid tumours from nonexposed patients. the gene CLIP2, located on chromosome 7q11.23.4 Based on these Several studies evidence that radiation exposure can induce copy findings, we aimed to investigate CLIP2 as a radiation biomarker in number alterations and deregulation of gene expressions with the PTC. However, one of the crucial requirements for a biomarker to potential of triggering carcinogenic processes, both of which was be used in epidemiological studies and risk modelling is its validity observed even at low doses of radiation.5–7 Common genetic in terms of sensitivity, specificity, reproducibility and biological

1Research Unit Radiation Cytogenetics, Helmholtz Zentrum München, German Research Center for Environmental Health GmbH, Neuherberg, Germany; 2Research Unit Analytical Pathology, Helmholtz Zentrum München, German Research Center for Environmental Health GmbH, Neuherberg, Germany; 3Institute of Endocrinology and Metabolism, National Academy of Medical Sciences of the Ukraine, Kiev, Ukraine; 4Institute of Radiation Protection, Helmholtz Zentrum München, German Research Center for Environmental Health GmbH, Neuherberg, Germany; 5Bundeswehr Institute of Radiobiology, Munich, Germany and 6Radiation Epidemiology Branch, Division of Cancer Epidemiology and Genetics, NCI, Bethesda, MD, USA. Correspondence: Dr J Hess, Research Unit Radiation Cytogenetics, Helmholtz Zentrum München, German Research Center for Environmental Health, Ingolstädter Landstr. 1, Neuherberg 85764, Germany. E-mail: [email protected] 7Clinical Cooperation Group ‘Personalized Radiotherapy of Head and Neck Cancer’, Helmholtz Zentrum München, German Research Center for Environmental Health GmbH, Neuherberg, Germany. Received 21 March 2014; revised 16 July 2014; accepted 9 August 2014; published online 6 October 2014 CLIP2 as radiation marker in PTC M Selmansberger et al 3918 plausibility. A further aspect for the integration into molecular sections were scanned and the generated images imported into epidemiology is the suitability of both a highly specific marker and the image viewer software Panoramic Viewer (3DHISTECH, an appropriate assay for its detection.14 In addition, a biomarker Budapest, Hungary). The staining intensities were visually scored may provide insights into the mechanisms of the disease it is a as follows: negative staining (score 0), weak staining (score 1), surrogate for. Therefore, this study also aimed to test whether a intermediate staining (score 2) and strong staining (score 3), as CLIP2 radiation-specific mRNA overexpression persists at the demonstrated in Figure 2. Detailed scoring criteria are outlined in protein level and to establish a reproducible workflow that allows the Materials and methods section. The entire classification a classification of PTCs from patients with unknown radiation workflow is illustrated in Figure 3. The resulting visual scores for history. Moreover, we intended to reconstruct the CLIP2 the 33 cases of the Genrisk-T cohort are shown in Table 1. A interactome from global transcriptome data that were derived statistically significant difference (two-sided Fisher’s exact test, from radiation-induced PTCs in order to clarify the role of CLIP2 in P-value = 0.005) between the exposed and nonexposed group was radiation-associated PTC carcinogenesis.15 also revealed using the visual scoring approach. Correlation analysis of the data from both evaluation approaches showed a strong correlation between digital analysis and visual scoring RESULTS − (Spearman’s correlation coefficients: 0.83, P-value: 1.63 × 10 6, CLIP2 protein expression is elevated in PTCs from patients Figure 4). In order to obtain an individual biomarker classification exposed to radioiodine fallout compared with a nonexposed for each case, cases with a visual score of 0 or 1 were considered control group as CLIP2 biomarker negative, whereas cases with visual score 3 In order to evaluate the expression of CLIP2 at the protein level were classified as CLIP2 biomarker positive. For a classification of in exposed and nonexposed cases of the Genrisk-T cohort, an intermediate CLIP2 staining (score 2), the genomic copy immunohistochemical staining with an antibody against CLIP2 number status (gained or not gained) of chromosomal band was performed in a highly standardized manner, followed by 7q11.23 (localization of CLIP2) was taken into account. Preferably, digital image analysis using the Definiens software (Definiens AG, data from array comparative genomic hybridization (array CGH) Munich, Germany). The obtained average marker staining were used. However, if array CGH data were not available, intensities within the analysed tumour regions (regions of interest) interphase fluorescence in situ hybridization analysis with a probe are listed in Supplementary Table 1. Statistical testing revealed specific for 7q11.23 was performed. Cases with a visual score of 2 significantly increased staining intensities (Mann–Whitney test, and a DNA gain of chromosomal band 7q11.23 were finally P-value = 0.0149) in PTC tissues from the exposed group as classified as CLIP2-positive, and those showing normal copy compared with the nonexposed group (Figure 1). Representative number of chromosomal band 7q11.23 were classified as CLIP2- immunohistochemically stained formalin-fixed, paraffin-embedded negative (Figure 3). (FFPE) tumour sections from nonexposed and exposed patients are shown in Figure 1, illustrating a radiation-associated over- Analysis of CLIP2 as a radiation biomarker in three independent expression of the CLIP2 protein. tumour cohorts In order to validate CLIP2 as a radiation biomarker in PTCs of Establishment of a workflow for an individual classification of the young patients, two independent tumour cohorts (Genrisk-T-PLUS CLIP2 expression and UkrAm) comprising in total 115 Ukrainian PTC cases were We simplified and standardized the whole procedure of CLIP2 analysed in addition to the discovery set Genrisk-T. Genrisk-T- protein expression measurement and established an approach for PLUS, which is an extension of the initial Genrisk-T cohort, scoring and classification of individual cases by visual inspection included exposed and nonexposed cases, whereas the set of PTC of digital immunohistochemistry (IHC) images. Subsequently, we cases from the UkrAm study included PTCs from exposed patients compared the results with the above-mentioned software-based only.16 The investigated cohorts are summarized in Table 2. All approach. For this purpose, the automated stained FFPE tissue PTC cases were individually scored for CLIP2 expression using our

Figure 1. Immunohistochemical staining of FFPE tumor sections from patients not exposed and exposed to radioiodine fallout using an antibody against CLIP2. A significantly increased CLIP2 expression (P = 0.0149) in exposed (n = 16, green boxplot) compared with nonexposed cases (n = 17, red boxplot) is shown at the protein level by IHC (middle). The marker staining intensities were evaluated by relative quantification using the Definiens software. The P-value was calculated using the Mann–Whitney test. Representative immunohistochemically stained FFPE sections from PTCs from nonexposed (left side; cases from top to bottom: UA0312, UA1328 and UA1208) and exposed (right side; cases from top to bottom: UA0905, UA0648 and UA0501) patients are shown.

Figure 2. Visual scoring of immunohistochemically stained FFPE tumour sections using an antibody against CLIP2. The marker staining intensities were evaluated by visual scoring 0–3. Two representative immunohistochemically stained papillary thyroid carcinoma cases with scores of 0, 1, 2 and 3, respectively, are shown from the left to the right. Image details of A-a and B-a (black frames) are shown below in A-b and B-b, respectively. standardized visual scoring approach. Figure 5 summarizes the with CLIP2) association of the network could be validated by resulting CLIP2 scores and the final classification into CLIP2- quantitative reverse transcriptase–PCR (qRT–PCR; Supplementary positive and -negative cases for all three independent tumour Table 2). The second neighbourhood (including the first neigh- cohorts. The results of single cases are listed in Supplementary bourhood, Supplementary Figure 2 and Supplementary Table 3) of Table 1. Gain of the chromosomal band 7q11.23 in score-2 cases CLIP2 contained 218 nodes and 1304 edges, including the gene was present in 0/9 Genrisk-T cases (0%), 4/11 Genrisk-T PLUS cases LMO3 that was shown to be associated with radiation dose in the (36%) and 7/18 UkrAm cases (39%). In Supplementary Figure 1, study by Abend et al.15 Pathway enrichment analysis of the array CGH profiles representing chromosome 7 from six UkrAm second neighbourhood revealed the significantly enriched cases are exemplarily shown. Following the inclusion of the pathways signalling by Nodal, transport of mature transcript to genomic copy number status of chromosomal band 7q11.23, we cytoplasm, Ras activation upon Ca2 influx through NMDA receptor obtained sensitivity rates of 75%, 75% and 72.4% for the Genrisk-T, and immune system. Genrisk-T-PLUS and UkrAm cohorts, respectively. The specificity rates were 82.4 and 57.1% for the Genrisk-T and Genrisk-T-PLUS DISCUSSION cohorts (Figures 5d–f). We continued and extended our previous study describing the Association of CLIP2 protein expression with patient and genomic radiation marker 7q11.23 that was exclusively detected histological data in PTCs from patients exposed to the Chernobyl radioiodine fallout and the radiation-associated CLIP2 overexpression at the The CLIP2 protein expression (IHC CLIP2 visual scores 0–3) was not mRNA level.4 Here, we confirmed CLIP2 overexpression in PTCs associated with age at exposure, sex, histological dominant from radiation-exposed patients at the protein level. Moreover, we pattern, presence of RET/PTC rearrangements or BRAF V600E developed a generally applicable workflow for the classification of mutations. CLIP2 protein expression was associated with exposure PTCs into CLIP2-positive or -negative cases and proposed CLIP2 as to radioiodine (P = 0.0015). Testing was applied to the merged a surrogate biomarker for radiation exposure in PTCs, allowing for clinical data on all three cohorts. subsequent integration of CLIP2 into epidemiological studies and thyroid cancer risk models. Furthermore, we reconstructed the CLIP2 gene regulatory network reconstruction CLIP2 interactome from published global mRNA expression data The global gene regulatory network based on global gene of the UkrAm cohort.15 expression data from 31 UkrAm cases consisted of 4746 nodes We were able to confirm the previously demonstrated (genes) and 29 842 edges (interactions between genes).15 The first radiation-associated CLIP2 mRNA expression at the protein level neighbourhood network of CLIP2, which was extracted from the in the same discovery cohort of cases (Figure 1). Within this global gene regulatory network, consisted of the seven genes cohort, the CLIP2 protein overexpression was present in 7q11.23 BAG2, CHST3, GOLM1, KIF3C, NEURL1, PPIL3 and RGS4 (Figure 6). gained cases. The observed variation of CLIP2 protein expression Furthermore, four of the CLIP2 first neighbourhood genes were in cases with normal copy number on chromosome 7q11.23 can associated with each other (KIF3C with BAG2 and RGS4, BAG2 with be explained by epigenetic regulations of CLIP2 expression, for RGS4 and NEURL1 with RGS4). All but one (interaction of GOLM1 example, by miRNA miR-16-5p (http://mirtarbase.mbc.nctu.edu.

Table 1. IHC CLIP2 visual scoring for PTC cases of the exposed and nonexposed groups (Genrisk-T cohort)

Score 1 Score 2 Score 3

Exposed (n = 16) 2 2 12 Nonexposed (n = 17) 7 7 3 Abbreviations: IHC, immunohistochemistry; PTC, papillary thyroid carcinoma.

Figure 4. Correlation between relative marker staining intensity (Definiens software) and visual scores. Good correlation (Spearman’s correlation coefficient rho = 0.83, P-value = 1.63 × 10 − 9)ofcomputer- based analysis (Definiens software) and visual scores of IHC CLIP2 stainings (scores 0–3; no case with score 0 within the Genrisk-T cohort) indicates consistency of analysis methods and justifies the transition from the analysis by image processing software to visual scoring evaluation.

Table 2. Number of radiation-exposed and nonexposed papillary thyroid carcinoma (PTC) cases in the three investigated cohorts

Number of cases in cohorts Genrisk-T Genrisk-T-PLUS UkrAm

Exposed 16 32 76 Nonexposed 17 7 —

digital image. Digital IHC slide images as generated by different slide scanner models might exhibit varying colour spectra because of different implementations in their hardware and software.17 Another variable is the manufacturer-dependent image processing software that might use diverging colour recognition algorithms, resulting in different relative staining intensities and intensity ranges within one data set. Therefore, because of the Figure 3. Schematic workflow for CLIP2 biomarker classification. influences of the above-mentioned parameters, the calculated average staining intensities generated by one lab are hardly tw), or other so far unknown epigenetic mechanisms like DNA reproducible by another. However, in order to use the CLIP2 methylation or histone modifications. In order to perform CLIP2 biomarker in molecular epidemiology, major requirements include typing of individual cases, we subsequently established a its reproducibility and the possibility to classify individual cases.14 simplified and standardized procedure. The motivation behind For this purpose, we established a generally applicable approach this effort was based on the fact that the continuous intensity for CLIP2 typing without the use of a digital analysis software. The values generated with a digital image analysis software strongly proposed approach for classifying individual cases in one out of depend on the used digital slide scanner, the applied quantifica- four categories is comparable to the procedures used in clinical tion algorithm and the regions of interest defined within the routine diagnostics, for example, HER2neu typing in breast

Figure 5. Stacked bar plots a, b, and c show the percentages of scores 0, 1, 2 and 3 of IHC staining intensities for the three different cohorts and their subgroups of exposed and nonexposed cases. Stacked bar plots d, e, and f show the classiﬁcation into biomarker-positive and biomarker-negative cases for the three different cohorts after inclusion of the genomic-level information (presence of gain on chromosomal band 7q11.23).

with the sensitivity of 75.0%, points to a high quality of the CLIP2 biomarker. The Genrisk-T-PLUS cohort was classified with the same sensitivity of 75.0% but a lower specificity of 57.1%, probably because of the low number of nonexposed cases in this cohort. However, because of the very low rate of sporadically occurring PTCs diagnosed at young age, the number of nonexposed cases within this cohort cannot be increased within a reasonable timeframe.19 The sensitivity rate of 72.4% within the UkrAm cohort of exposed cases is similar to the rates determined in the Genrisk- T and Genrisk-T-PLUS cohort. Moreover, the sensitivity rates for Genrisk-T, Genrisk-T-PLUS and UkrAm are in good accordance with the estimated proportion of patients with radiation-induced Figure 6. First neighbourhood network of CLIP2. De novo gene PTC in the exposed cohorts. Hess et al.4 estimated that ∼ 85% of regulatory network reconstruction was performed using the PTCs from the exposed group of the Genrisk-T cohort were indeed GeneNet method. All edges were independently validated by – radiation induced. Compared with the Genrisk-T cohort (mean age qRT PCR (high correlation between the expressions of connected at exposure 2 years; mean age at operation 16 years), patients genes), except the association between GOLM1 and CLIP2. from the UkrAm cohort were in average older at the time of exposure (8 years) and at the time of operation (25 years). The cancer.18 For CLIP2 typing we particularly aimed at a simplification proportion of radiation-induced cases in the UkrAm cohort was while ensuring crucial aspects of biomarker detection such as determined using the estimated average thyroid dose of 0.65 Gy reproducibility, sensitivity and specificity. The visual scoring results and an excess relative risk of 5 Gy − 1 for the screening prevalence almost perfectly reflected the initial software-derived results, cohort and 1.91 Gy − 1 for the incidence cases (second to fourth justifying the proposed simplification (Figure 4). Beyond this screening).20,21 Consequently, the estimated proportion of methodological optimization, we validated a radiation-associated patients with radiation induced PTC in the UkrAm cohort was CLIP2 overexpression in two additional independent tumour 55–75%. The sensitivity rates for the CLIP2 marker of 75% and 72% cohorts (Genrisk-T-PLUS and UkrAm). The very similar sensitivity for Genrisk-T and UkrAm, respectively, are likely to reflect the rates between 72.4 and 75.0% in all three cohorts demonstrate estimated frequencies of radiation-induced PTCs and suggest a both the robustness of the approach and the ability to detect a very sensitive detection of radiation-induced cases. high number of true positive cases (Figure 5). The specificity in the The radiation-specific overexpression of CLIP2 in PTCs on both Genrisk-T cohort of 82.4% is particularly high and, in conjunction the mRNA and the protein levels strongly suggests an important

© 2015 Macmillan Publishers Limited Oncogene (2015) 3917 – 3925 CLIP2 as radiation marker in PTC M Selmansberger et al 3922 role of CLIP2 in radiation-induced carcinogenesis of PTC. In the complex.48 Beside their function in microtubule-dependent published literature, CLIP2 is mainly known to be associated transport, kinesin motor proteins also contribute to the organiza- with the Williams–Beuren syndrome because of its genomic tion and movement of spindle poles and chromosomes during localization within the Williams–Beuren syndrome hemizygous mitosis.49–51 Therefore, KIF3C might represent an essential mitotic deletion.22,23 Amplifications of CLIP2 have been detected in component required for accurate cell division including accurate glioblastomas and colorectal carcinomas.24,25 CLIPs are cytoplas- chromosome segregation—prerequisites for genomic stability. An matic linker proteins binding to the ends of growing microtubules interaction of KIF3C with CLIP2 would indicate a potential role of and are thereby involved in the dynamics regulation of the CLIP2 in chromosome segregation and cell division as it has cytoskeletal network including the localization of the dynein– already been shown and discussed above for CLIP1. Genomic dynactin complex.26,27 The latter, in turn, is involved in various instability is a well-established phenotype after irradiation and processes such as spindle organization, chromosome alignment might be a critical step in the radiation-associated and chromosome segregation in cell division.28 Moreover, CLIP2 carcinogenesis.52–55 An effect of CLIP2 on genomic instability shows high structural similarities to the protein CLIP1 and shares could therefore contribute to radiation-induced carcinogenesis of conserved protein domains like cytoskeleton-associated protein PTC. Moreover, the identification of CLIP2 interacting genes opens glycine rich and structural maintenance of chromosomes that is the possibility to investigate their role as additional biomarkers in linked to chromosome segregation and cell division.29,30 CLIP1 is radiation-induced thyroid carcinogenesis. essential in the G2/M transition, facilitates the formation of In conclusion, this study provides the validation of the recently kinetochore–microtubule attachments during mitosis and inter- published radiation marker CLIP2 at the protein level in radiation- – acts directly with the dynein–dynactin complex.31 34 For CLIP2 an associated PTCs from independent cohorts. We established a indirect connection to the dynein–dynactin complex via BICD2 is standardized procedure for CLIP2 typing, an essential step in proposed.35,36 Despite these published links, a detailed knowledge integrating a molecular marker into epidemiological studies. about the function of CLIP2 is quite limited. Finally, analysis of the CLIP2 interactome suggests the involve- To gain knowledge about the functional role of CLIP2 in ment of CLIP2 in the fundamental carcinogenic processes radiation-associated PTC, we used the reconstructed gene apoptosis, MAPK signalling and genomic instability, indicating a regulatory network inferred from microarray transcriptome data functional role of CLIP2 in the carcinogenesis of radiation- and extracted the putative CLIP2 interaction partners from this associated PTC. network.15 The global mRNA expression data used for GRN reconstruction were previously published by Abend et al.15 and were generated from PTC tissue samples from 31 patients of MATERIALS AND METHODS the UkrAm cohort.15 A subgroup (n = 15) of UkrAm cases of the Patient data and tumour tissues transcriptome data set was also subjected to CLIP2 typing in this Tissue samples from 124 PTCs that developed in young patients (0.1–17 study. Thus, the reconstructed GRN was partly conducted on the years of age at exposure, born before 26 April 1986) after exposure to same cases that were used for CLIP2 biomarker typing. We assume radioiodine fallout as a consequence of the Chernobyl reactor accident as that this network represents the CLIP2 interactome, that is, its well as 24 sporadic PTCs were obtained from the CTB. The patients were direct or indirect interaction partners. The identified six first residents in one of the following oblasts of Ukraine: Cherkassy, Chernigov, network neighbours of CLIP2 (BAG2, CHST3, KIF3C, NEURL1, RGS4 Kiev (including Pripyat city), Rovno, Sumy or Zhytomyr. Sporadic PTC cases and PPIL3), which were validated by qRT–PCR on RNA from FFPE from patients born after 1 January 1987, and therefore not exposed to radioiodine fallout from the Chernobyl accident, were matched on sections from the same cohort, are known to be involved in residency, age at operation and sex to the exposed PTC cases. Pathological fundamental carcinogenic processes, indicating a functional role diagnosis was performed at the Laboratory of Morphology of Endocrine of CLIP2 in the carcinogenesis of radiation-associated PTC. The System (IEM, Kiev, Ukraine) by two pathologists (LZ/TB) and reviewed by published literature reveals that these genes are likely to be the CTB Pathology Panel.56 All tumours were diagnosed as PTCs. The linked to the hallmarks of cancer resisting cell death, sustaining dominant histological patterns (follicular/papillary/solid) of the studied proliferative signalling and genome instability and mutation.37 BAG2 sections were determined (by LZ/TB). An overview of the investigated and NEURL1 are involved in apoptosis and might therefore play a tumour cohorts is given in Table 2. The patients’ individual data are listed role in evading cell death by epithelial thyroid cancer cells.37–39 in Supplementary Table 1. We analysed a discovery cohort consisting of 33 ‘ ’ BAG2 as well as RGS4 are known to play a role in thyroid cancer PTC cases (16 exposed and 17 nonexposed cases, the so-called Genrisk-T ) and are also known to be involved in the MAPK signalling and two validation cohorts consisting of 39 PTCs (32 exposed and 7 40,41 nonexposed cases, so-called ‘Genrisk-T-PLUS’) and 76 PTCs (exposed cases pathway. The importance of the MAPK pathway has been well only, so-called ‘UkrAm’), respectively. FFPE tumour tissue sections of all established in thyroid carcinogenesis. The MAPK pathway is cases and tumour DNA isolated from fresh-frozen tissue of the Genrisk-T frequently constitutively activated in PTCs by genetic alterations and Genrisk-T-PLUS cases were provided by the CTB. Tumour DNA from such as rearrangements of the RET gene (RET/PTC), TRK or BRAF or the UkrAm cases was isolated from FFPE tissue sections using the Qiagen by point mutations of the BRAF and RAS genes as well as by the AllPrep DNA/RNA FFPE Kit (Qiagen, Hilden, Gemany). Tumour RNA was recently discovered kinase fusion oncogenes ETV6–NTRK3.13,42,43 isolated using the Qiagen RNeasy FFPE Kit. RET/PTC1 and RET/PTC3 rearrangements, as well as BRAFV600E mutation status, were determined The aforementioned NEURL1 is known to be involved in the Notch 4 pathway that in turn is activated by MAPK signalling in PTC.38,44 as described previously. Furthermore, the subsequent pathway enrichment analysis including all first and second CLIP2 neighbourhood genes Immunohistochemistry revealed the significantly enriched pathways Ras activation upon Immunohistochemical staining of FFPE tumour tissue sections was Ca2 influx through NMDA receptor and signalling by Nodal. Both are performed using a primary antibody against CLIP2 (HPA020430; Sigma also connected to the MAPK pathway, the latter of which by a Prestige Antibodies, St Louis, MO, USA). The antibody was selected from molecular cross-talk between the aforementioned Notch pathway the ‘The Human Protein Atlas’ database that comprises information 45–47 about the antibody specificity and staining patterns.57,58 Antibody and Nodal signalling. Activation of the MAPK pathway plays a fi fundamental role in the regulation of cell proliferation, linking the speci city was validated by western blot analysis with protein lysates from a CLIP2-expressing cell line and a CLIP2 small interfering RNA function of the assumed CLIP2 interacting genes to the cancer fi (Ambion, Carlsbad, CA, USA, Silencer Select ID: s14847) knockdown cell line hallmark sustaining proliferative signalling. Moreover, the CLIP2 rst (Supplementary Figure 3). Primary antibody was used in a dilution of 1:100 neighbour KIF3C suggests a connection to the cancer hallmark in the automated staining instrument Discovery XT (Roche, Ventana, genome instability & mutation. The motor protein KIF3C belongs to Tucson, AZ, USA) and Discovery-Universal (Ventana) was used as secondary the kinesin family and is part of a microtubule-associated antibody. Signal detection was performed using peroxidase-DAB-

Oncogene (2015) 3917 – 3925 © 2015 Macmillan Publishers Limited CLIP2 as radiation marker in PTC M Selmansberger et al 3923 (diaminobenzidine)-MAP chemistry (Roche, Ventana). The stained tissue thereby providing a much more accurate estimate of pair-wise correlations sections were fixed in an ethanol series and coated by a coverslip before compared with traditional correlation approaches. The approach imple- scanning at × 20 objective magnification with a digital slide scanning mented in GeneNet is particularly tailored for ‘large-p small-n’ situations in system (Mirax Desk, Carl Zeiss MicroImaging, Jena, Germany). The resulting which the number of measurements is much higher than the number of staining was confirmed by a pathologist (AW). samples analysed. For the definition of edges (that is, connections) between two genes we used a considerably stringent probability cutoff of Image analysis of immunohistochemistry 0.988. From the resulting global network, the CLIP2-centred first Digital image analysis. The marker staining intensities were evaluated by neighbourhood network (direct neighbours) and second neighbourhood fi relative quantification using digital image analysis platform DefiniensTis- (neighbours of rst neighbours of CLIP2) were extracted. For pathway sueStudio (Definiens AG). For this purpose, the digital slide images were enrichment analysis, Reactome pathway gene sets (674) were downloaded 63 imported into the image analysis software using the tissue portal from GSEA. The gene list building the first and second neighbourhood (DefiniensTissueStudio). In the first step, regions of interest, that is, tumour was matched to the gene sets and tested for overrepresentation by two- areas, were defined. In order to detect and quantify stained tissue areas, sided Fisher’s exact test with an assumed number of total human genes of a continuous spectrum of brown staining intensity in relative units 19 104 (all HUGO Gene Nomenclature Committee (HGNC) annotated (0.00–3.00) was obtained using predefined algorithm and optimized genes).64 The P-values were corrected for multiple testing errors using the settings. Finally, the developed image analysis was automatically applied Benjamini–Hochberg correction.65 False discovery rates of o0.3 were to all digital images in a batch process analysing the relevant regions of considered statistically significant. interest. Quantitative RT–PCR Visual scoring classification. The visual scoring classification was carried out by two independent observers (JH/MS) in a blinded manner with Reverse transcription of RNA was performed using the VILO SuperScript respect to the PTC exposure status. Only epithelial tumour cells were Reverse Transcription Kit (Life Technologies, Carlsbad, CA, USA). The evaluated. Tumour stroma and infiltrations, such as lymphocytes, were not qRT–PCR reactions (10 μl) were carried out in duplicates in a ViiA 7 Real considered. The tumour area with the most pronounced IHC staining was Time PCR System in combination with the ViiA 7 Software v1.2.2 (Life used for scoring and classified into one of the four staining categories: Technologies). TaqMan gene expression assays (Life Technologies) negative staining (score 0), weak staining (score 1), intermediate staining detecting the following genes were used to validate the first neighbour- (score 2) and strong staining (score 3). Each case was independently hood network (see previous section) of CLIP2: CLIP2 (Hs00185593_m1), evaluated three times in a blinded scenario by each of the two observers. BAG2 (Hs00188716_m1), CHST3 (Hs00427946_m1), RGS4 (Hs01111690_g1), Thereby, six scores were obtained for each individual case. Subsequently, NEURL1 (Hs00907830_m1), KIF3C (Hs01547426_m1), PPIL3 (Hs00368985_m1) the modus (that is, the most frequent score) of all six scores was taken as a and GOLM1 (Hs00895845_m1). Assays detecting the genes RPL30 consensus result. In case of a bimodal distribution (that is, two different (Hs01066167_g1) and PGK1 (Hs99999906_m1) were used for the purpose scores with the same frequencies), the higher value was taken. If the six of endogenous normalization. Relative expression levels were calculated scores differed more than one scoring level, this particular case was using the ΔΔCt method.66 The expression levels of all eight genes excluded from further analysis. (including CLIP2) from the CLIP2 first neighbourhood network were determined in five PTC samples of the UkrAm cohort. Correlations of the Statistical analysis. The average marker staining intensities (obtained by genes were calculated based on the obtained expression values using fi the De niens software) of the FFPE PTC tissue sections from the exposed Spearman’s correlation method. Correlations were considered as ‘high’ and fi and nonexposed tumour groups were tested for statistical signi cant thus the association between the network genes was validated if the differences using the Mann–Whitney test (R base function wilxox.test with Spearman’s correlation coefficient was 40.6. ‘paired’ option set to false). Correlation of the continuous values (Definiens) with the visually assessed scores (0/1/2/3) was tested using Spearman’s correlation method (R base function cor.test). Possible associations of the ABBREVIATIONS obtained CLIP2 visual scores with clinicopathological and patient data were statistically tested using Fisher’s exact test (R base function fisher. CGH, comparative genomic hybridization; CTB, Chernobyl tissue test). P-values of o0.05 were considered statistically significant. bank; FFPE, formalin-fixed, paraffin-embedded; GRN, gene regulatory network; IHC, immunohistochemistry; PTC, papillary Genomic copy number typing thyroid carcinoma. In order to detect copy number aberrations on chromosome 7q11, array CGH analysis was performed using Agilent (Santa Clara, CA, USA) CONFLICT OF INTEREST 60k or 180k (AMADID 252192/252206) CGH microarrays as described previously.4 Alternatively, interphase fluorescence in situ hybridization The authors declare no conflict of interest. analysis was performed on FFPE tissue sections as described previously.4 ACKNOWLEDGEMENTS De novo reconstruction of the CLIP2 gene regulatory network using mRNA expression microarray data We thank the International Pathology Panel of the Chernobyl Tissue Bank for confirmation of diagnosis: Professors A Abrosimov, TI Bogdanova, G Fadda, G Hant, We used published global mRNA expression data from 31 PTC cases of the V LiVolsi, J Rosai and ED Williams; The Chernobyl Tissue Bank for collection of thyroid UkrAm cohort for the reconstruction of the CLIP2 GRN.15 Raw data import, tissue samples; Professor G Thomas for establishing the matched Genrisk-T cohort; filtering and normalization (preprocessing) of the data were carried out Dr Peter Jacob for discussion and determination of the proportion of radiation- within the statistical programming framework R using the Bioconductor package limma.59–61 In order to solely consider genes that are likely to play induced tumours among the exposed cases in the UkrAm cohort; U Buchholz, fl a role within the GRN, only probes binding to transcripts with curated C Innerlohinger, E Konhäuser, CM P üger and A Selmaier for technical support; and RefSeq records were considered for the analysis. Replicated expressions H Braselmann for mathematical/statistical support. This study was supported by the were summarized using the Limma function ‘summarize.probes’. The European Commission, EpiRadBio project, FP7 Grant No. 269553 and in part by the resulting matrix with log2-normalized expression values (13 662 genes and European Commission, DoReMi project, Grant No. 249689. 31 samples) was subjected to de novo network reconstruction using the Bioconductor package GeneNet.62 It is generally assumed that genes with a strong linear dependency in their expression patterns, and thus a high REFERENCES correlation between them, do either directly or indirectly interact with 1 Cardis E, Hatch M. The Chernobyl accident-an epidemiological perspective. Clin each other. The GeneNet package allows the reconstruction of GRNs based Oncol 2011; 23: 251–260. on pairwise partial correlation of gene expressions. Partial correlation is a 2 Tuttle RM, Vaisman F, Tronko MD. Clinical presentation and clinical outcomes in measure of the degree of association between two gene expression Chernobyl-related paediatric thyroid cancers: what do we know now? What can vectors after removing the influence of all other genes in the data set and we expect in the future? Clin Oncol 2011; 23: 268–275.

© 2015 Macmillan Publishers Limited Oncogene (2015) 3917 – 3925 CLIP2 as radiation marker in PTC M Selmansberger et al 3924 3 Cardis E, Howe G, Ron E, Bebeshko V, Bogdanova T, Bouville A et al. Cancer genes in chromosomal- and microsatellite-unstable sporadic colorectal carcino- consequences of the Chernobyl accident: 20 years on. J Radiol Prot 2006; 26: mas. JMolMed2007; 85:293–304. 127–140. 26 Hoogenraad CC, Akhmanova A, Grosveld F, De Zeeuw CI, Galjart N. Functional 4 Hess J, Thomas G, Braselmann H, Bauer V, Bogdanova T, Wienberg J et al. Gain of analysis of CLIP-115 and its binding to microtubules. J Cell Sci 2000; 113(Pt 12): chromosome band 7q11 in papillary thyroid carcinomas of young patients is 2285–2297. associated with exposure to low-dose irradiation. Proc Natl Acad Sci USA 2011; 27 Galjart N. CLIPs and CLASPs and cellular dynamics. Nat Rev Mol Cell Biol 2005; 6: 108: 9595–9600. 487–498. 5 Iizuka D, Imaoka T, Takabatake T, Nishimura M, Kakinuma S, Nishimura Y et al. 28 Karki S, Holzbaur EL. Cytoplasmic dynein and dynactin in cell division and intra- DNA copy number aberrations and disruption of the p16Ink4a/Rb pathway in cellular transport. Curr Opin Cell Biol 1999; 11:45–53. radiation-induced and spontaneous rat mammary carcinomas. Radiat Res 2010; 29 Hoogenraad CC, Akhmanova A, Galjart N, De Zeeuw CI. LIMK1 and CLIP-115: 174:206–215. linking cytoskeletal defects to Williams syndrome. Bioessays 2004; 26:141–150. 6 Ishida Y, Takabatake T, Kakinuma S, Doi K, Yamauchi K, Kaminishi M et al. Genomic 30 Marchler-Bauer A, Anderson JB, Chitsaz F, Derbyshire MK, DeWeese-Scott C, and gene expression signatures of radiation in medulloblastomas after low-dose Fong JH et al. CDD: specific functional annotation with the Conserved Domain irradiation in Ptch1 heterozygous mice. Carcinogenesis 2010; 31: 1694–1701. Database. Nucleic Acids Res 2009; 37: D205–D210. 7 Mullenders L, Atkinson M, Paretzke H, Sabatier L, Bouffler S. Assessing cancer risks 31 Yang X, Li H, Liu XS, Deng A, Liu X. Cdc2-mediated phosphorylation of CLIP-170 is of low-dose radiation. Nat Rev Cancer 2009; 9:596–604. essential for its inhibition of centrosome reduplication. J Biol Chem 2009; 284: 8 Klugbauer S, Lengfelder E, Demidchik EP, Rabes HM. High prevalence of RET 28775–28782. rearrangement in thyroid tumors of children from Belarus after the Chernobyl 32 Wieland G, Orthaus S, Ohndorf S, Diekmann S, Hemmerich P. Functional reactor accident. Oncogene 1995; 11:2459–2467. complementation of human centromere protein A (CENP-A) by Cse4p from 9 Hamatani K, Eguchi H, Ito R, Mukai M, Takahashi K, Taga M et al. RET/PTC Saccharomyces cerevisiae. Mol Cell Biol 2004; 24: 6620–6630. rearrangements preferentially occurred in papillary thyroid cancer among 33 Tanenbaum ME, Galjart N, van Vugt MA, Medema RH. CLIP-170 facilitates the atomic bomb survivors exposed to high radiation dose. Cancer Res 2008; 68: formation of kinetochore-microtubule attachments. EMBO J 2006; 25:45–57. 7176–7182. 34 Lansbergen G, Komarova Y, Modesti M, Wyman C, Hoogenraad CC, Goodson HV 10 Leeman-Neill RJ, Brenner AV, Little MP, Bogdanova TI, Hatch M, Zurnadzy LY et al. et al. Conformational changes in CLIP-170 regulate its binding to microtubules RET/PTC and PAX8/PPARgamma chromosomal rearrangements in post-Chernobyl and dynactin localization. J Cell Biol 2004; 166: 1003–1014. thyroid cancer and their association with iodine-131 radiation dose and other 35 Hoogenraad CC, Wulf P, Schiefermeier N, Stepanova T, Galjart N, Small JV et al. characteristics. Cancer 2013; 119: 1792–1799. Bicaudal D induces selective dynein-mediated microtubule minus end-directed 11 Tuttle RM, Lukes Y, Onstad L, Lushnikov E, Abrosimov A, Troshin V et al. ret/PTC transport. EMBO J 2003; 22: 6004–6015. activation is not associated with individual radiation dose estimates in a pilot 36 Fumoto K, Hoogenraad CC, Kikuchi A. GSK-3beta-regulated interaction of BICD study of neoplastic thyroid nodules arising in Russian children and adults exposed with dynein is involved in microtubule anchorage at centrosome. EMBO J 2006; to Chernobyl fallout. Thyroid 2008; 18:839–846. 25: 5670–5682. 12 Nikiforov YE, Rowland JM, Bove KE, Monforte-Munoz H, Fagin JA. Distinct pattern 37 Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011; of ret oncogene rearrangements in morphological variants of radiation-induced 144:646–674. and sporadic thyroid papillary carcinomas in children. Cancer Res 1997; 57: 38 Teider N, Scott DK, Neiss A, Weeraratne SD, Amani VM, Wang Y et al. Neuralized1 1690–1694. causes apoptosis and downregulates Notch target genes in medulloblastoma. 13 Ricarte-Filho JC, Li S, Garcia-Rendueles ME, Montero-Conde C, Voza F, Knauf JA Neuro Oncol 2010; 12: 1244–1256. et al. Identification of kinase fusion oncogenes in post-Chernobyl radiation- 39 Wang HQ, Zhang HY, Hao FJ, Meng X, Guan Y, Du ZX. Induction of BAG2 protein induced thyroid cancers. J Clin Invest 2013; 123: 4935–4944. during proteasome inhibitor-induced apoptosis in thyroid carcinoma cells. Br J 14 Pernot E, Hall J, Baatout S, Benotmane MA, Blanchardon E, Bouffler S et al. Ionizing Pharmacol 2008; 155: 655–660. radiation biomarkers for potential use in epidemiological studies. Mutat Res 2012; 40 Arndt V, Daniel C, Nastainczyk W, Alberti S, Hohfeld J. BAG-2 acts as an inhibitor of 751: 258–286. the chaperone-associated ubiquitin ligase CHIP. Mol Biol Cell 2005; 16: 5891–5900. 15 Abend M, Pfeiffer RM, Ruf C, Hatch M, Bogdanova TI, Tronko MD et al. Iodine-131 41 Druey KM, Blumer KJ, Kang VH, Kehrl JH. Inhibition of G-protein-mediated MAP dose dependent gene expression in thyroid cancers and corresponding normal kinase activation by a new mammalian gene family. Nature 1996; 379: 742–746. tissues following the Chernobyl accident. PLoS ONE 2012; 7: e39103. 42 Nikiforov YE. Thyroid carcinoma: molecular pathways and therapeutic targets. 16 Tronko M, Mabuchi K, Bogdanova T, Hatch M, Likhtarev I, Bouville A et al. Thyroid Mod Pathol 2008; 21(Suppl 2): S37–S43. cancer in Ukraine after the Chernobyl accident (in the framework of the Ukraine- 43 Xing M. Molecular pathogenesis and mechanisms of thyroid cancer. Nat Rev US Thyroid Project). J Radiol Prot 2012; 32: N65–N69. Cancer 2013; 13:184–199. 17 Rojo MG, Bueno G, Slodkowska J. Review of imaging solutions for integrated 44 Yamashita AS, Geraldo MV, Fuziwara CS, Kulcsar MA, Friguglietti CU, da Costa RB quantitative immunohistochemistry in the Pathology daily practice. Folia His- et al. Notch pathway is activated by MAPK signaling and influences papillary tochem Cytobiol 2009; 47: 349–354. thyroid cancer proliferation. Transl Oncol 2013; 6:197–205. 18 Varga Z, Noske A, Ramach C, Padberg B, Moch H. Assessment of HER2 status in 45 Strizzi L, Hardy KM, Seftor EA, Costa FF, Kirschmann DA, Seftor RE et al. breast cancer: overall positivity rate and accuracy by fluorescence in situ hybri- Development and cancer: at the crossroads of Nodal and Notch signaling. Cancer dization and immunohistochemistry in a single institution over 12 years: a quality Res 2009; 69: 7131–7134. control study. BMC Cancer 2013; 13: 615. 46 Grewal T, Koese M, Tebar F, Enrich C. Differential regulation of RasGAPs in cancer. 19 Williams D. Cancer after nuclear fallout: lessons from the Chernobyl accident. Nat Genes Cancer 2011; 2:288–297. Rev Cancer 2002; 2:543–549. 47 Xia Z, Dudek H, Miranti CK, Greenberg ME. Calcium influx via the NMDA receptor 20 Tronko MD, Howe GR, Bogdanova TI, Bouville AC, Epstein OV, Brill AB et al. induces immediate early gene transcription by a MAP kinase/ERK-dependent A cohort study of thyroid cancer and other thyroid diseases after the chornobyl mechanism. J Neurosci 1996; 16: 5425–5436. accident: thyroid cancer in Ukraine detected during first screening. J Natl Cancer 48 Muresan V, Abramson T, Lyass A, Winter D, Porro E, Hong F et al. KIF3C and KIF3A Inst 2006; 98:897–903. form a novel neuronal heteromeric kinesin that associates with membrane 21 Brenner AV, Tronko MD, Hatch M, Bogdanova TI, Oliynik VA, Lubin JH et al. I-131 vesicles. Mol Biol Cell 1998; 9:637–652. dose response for incident thyroid cancers in Ukraine related to the Chornobyl 49 Goldstein LS. Kinesin molecular motors: transport pathways, receptors, and accident. Environ Health Perspect 2011; 119:933–939. human disease. Proc Natl Acad Sci USA 2001; 98: 6999–7003. 22 Hoogenraad CC, Koekkoek B, Akhmanova A, Krugers H, Dortland B, Miedema M 50 Mandelkow E, Mandelkow EM. Kinesin motors and disease. Trends Cell Biol 2002; et al. Targeted mutation of Cyln2 in the Williams syndrome critical region links 12:585–591. CLIP-115 haploinsufficiency to neurodevelopmental abnormalities in mice. Nat 51 De S, Cipriano R, Jackson MW, Stark GR. Overexpression of kinesins mediates Genet 2002; 32: 116–127. docetaxel resistance in breast cancer cells. Cancer Res 2009; 69: 8035–8042. 23 Ferrero GB, Howald C, Micale L, Biamino E, Augello B, Fusco C et al. An atypical 52 Little MP, Heidenreich WF, Moolgavkar SH, Schollnberger H, Thomas DC. Systems 7q11.23 deletion in a normal IQ Williams-Beuren syndrome patient. Eur J Hum biological and mechanistic modelling of radiation-induced cancer. Radiat Environ Genet 2010; 18:33–38. Biophys 2008; 47:39–47. 24 Suzuki T, Maruno M, Wada K, Kagawa N, Fujimoto Y, Hashimoto N et al. Genetic 53 Huang L, Snyder AR, Morgan WF. Radiation-induced genomic instability and its analysis of human glioblastomas using a genomic microarray system. Brain Tumor implications for radiation carcinogenesis. Oncogene 2003; 22: 5848–5854. Pathol 2004; 21:27–34. 54 Morgan WF. Non-targeted and delayed effects of exposure to ionizing radiation: I. 25 Lassmann S, Weis R, Makowiec F, Roth J, Danciu M, Hopt U et al. Array CGH Radiation-induced genomic instability and bystander effects in vitro. Radiat Res identifies distinct DNA copy number profiles of oncogenes and tumor suppressor 2003; 159:567–580.

Oncogene (2015) 3917 – 3925 © 2015 Macmillan Publishers Limited CLIP2 as radiation marker in PTC M Selmansberger et al 3925 55 Morgan WF. Non-targeted and delayed effects of exposure to ionizing radiation: 61 Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S et al. II. Radiation-induced genomic instability and bystander effects in vivo, clastogenic Bioconductor: open software development for computational biology and factors and transgenerational effects. Radiat Res 2003; 159:581–596. bioinformatics. Genome Biol 2004; 5:R80. 56 Williams ED. Guest Editorial: two proposals regarding the terminology of 62 Opgen-Rhein R, Strimmer K. From correlation to causation networks: a simple thyroid tumors. Int J Surg Pathol 2000; 8: 181–183. approximate learning algorithm and its application to high-dimensional plant 57 Uhlen M, Bjorling E, Agaton C, Szigyarto CA, Amini B, Andersen E et al. A human gene expression data. BMC Syst Biol 2007; 1:37. protein atlas for normal and cancer tissues based on antibody proteomics. Mol 63 Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA et al. Cell Proteom 2005; 4: 1920–1932. Gene set enrichment analysis: a knowledge-based approach for interpreting 58 Uhlen M, Oksvold P, Fagerberg L, Lundberg E, Jonasson K, Forsberg M et al. genome-wide expression proﬁles. Proc Natl Acad Sci USA 2005; 102: 15545–15550. Towards a knowledge-based Human Protein Atlas. Nat Biotechnol 2010; 28: 64 Gray KA, Daugherty LC, Gordon SM, Seal RL, Wright MW, Bruford EA. 1248–1250. Genenames.org: the HGNC resources in 2013. Nucleic Acids Res 2013; 41: 59 Team RDC. R: A Language and Environment for Statistical Computing. D545–D552. R Foundation for Statistical Computing, 2013. 65 Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and 60 Smyth GK. Limma: linear models for microarray data. In: Gentleman R, powerful approach to multiple testing. J R Stat So 1995; 57: 289–300. Carey V, Dudoit S, Irizarry R, Huber W (eds) Bioinformatics and Computational 66 Livak KJ, Schmittgen TD. Analysis of relative gene expression data using Biology Solutions Using {R} and Bioconductor. Springer: New York, 2005, pp real-time quantitative PCR and the 2(-Delta Delta C(T)) method. Methods 2001; 25: 397–420. 402–408.

Supplementary Information accompanies this paper on the Oncogene website (http://www.nature.com/onc)