Impact of medical imaging on the epigenome – low-dose exposure in the course of computed tomography does not induce detectable changes of DNA-methylation profiles in peripheral blood cells

Abstract

Background

Computed tomography (CT) is a main contributor to artificial low-dose exposure. Understanding the biological effects induced by CT exposure and their dependency on the characteristics of photon spectra is essential for knowledge-driven risk assessment. In a previous gene expression study, we have identified upregulation of AEN, BAX, DDB2, EDA2R and FDXR after ex vivo exposure with single-energy CT and dual-energy CT (DECT). In this study, we focused on CT-induced changes of DNA methylation. This epigenetic modification of DNA is a central regulator of gene expression and instrumental in preserving genome integrity. Previous studies reported focal hypermethylation and global hypomethylation after exposure with doses above 100 mSv, however, the effect of low dose exposure on DNA methylation is hardly explored.

Materials and methods

DNA was isolated from peripheral blood of three healthy individuals 6 h after ex vivo exposition to single-energy (80 kV and 150 kV) and DECT (80 kV/Sn150 kV) with a calculated effective dose of 7.0 ± 0.08 mSv. The experimental setting was identical to the one used in our previous gene expression study enabling a direct comparison of gene expression results with changes of DNA methylation identified in this study. DNA methylation was analyzed by high-throughput sequencing of bisulfite-treated DNA targeted methylation sequencing.

Results

Unsupervised hierarchical clustering based on DNA methylation profiles of all samples created three distinct clusters. Formation of these three clusters was solely determined by the origin of samples, indicating the absence of prominent irradiation-associated changes of DNA methylation. In line with this observation, inter-individual comparison of non-irradiated samples revealed 1163, 1224 and 4550 significant differentially methylated regions (DMRs), respectively, whereas the pairwise comparison of irradiated and non-irradiated samples failed to identify irradiation-induced DMRs in any of the three probands. This even applied to the genomic regions harboring AEN, BAX, DDB2, EDA2R and FDXR, the five genes known to be upregulated by CT exposure.

Conclusions

CT exposure with various photon spectra did not result in detectable changes of DNA methylation. However, minor effects in a subpopulation of irradiated cells cannot be ruled out. Thus, future studies with extended observation intervals are needed to investigate DNA methylation changes that are induced by indirect effects at later points of time or become detectable by clonal expansion of affected cells. Moreover, our data suggest that DNA methylation analysis is less sensitive in detecting immediate effects of low-dose irradiation when compared to gene expression analysis.

Keywords:

• Computed tomography (CT)
• genomics
• low dose effects
• epigenomics

Introduction

Radiological examinations are the main source of artificial low-dose exposure. The greatest fraction of it is caused by computed tomography (CT) examinations, which meanwhile account for 63% of collective effective dose in the context of radiological investigations in the US and 67% in Germany, respectively (Charles 2008; Nekolla et al. 2017; Mettler et al. 2020). From 2007 to 2016 alone, the frequency of CT examinations has risen by 45% in Germany. Similar trends have been reported for other industrial nations (Organisation for Economic Co-operation and Development 2018). In part, the increased usage of CT is owed to technical improvements such as the introduction of dual-energy CT (DECT), which takes advantage of combining two different tube voltages. This results in photons with different adsorption properties and thereby an enhanced ability to discriminate tissue compositions. Although the risk associated with a single CT examination can be expected to be very low (Hall and Brenner 2008), considering the stochastic nature of low-dose effects and the large number of CT examinations the overall population risk cannot be ignored. Understanding the biological processes that shape the cellular response to low dose irradiation is key to knowledge-driven risk assessment.

DNA methylation is an epigenetic modification that is involved in a broad spectrum of biological processes ranging from embryogenesis (Okano et al. 1999), X-inactivation (Riggs 1975), genomic imprinting (Li et al. 1993), cell differentiation (Jones 2012) and suppression of transposable elements (Slotkin and Martienssen 2007) to learning and memory (Lister et al. 2013). In order to fulfill its role in these processes, DNA methylation is adaptable in a dynamic and regionally controlled fashion in response to intrinsic and environmental cues. Against this background, several studies have been conducted to elucidate the impact of ionizing radiation on DNA methylation (Kalinich et al. 1989; Pogribny et al. 2004; Koturbash et al. 2005; Pogribny et al. 2005; Kumar et al. 2011). Some of these studies had a gene centered view on the methylome, focusing on the DNA methylation state of CpG islands as a predictor of transcriptional activity (Kalinich et al. 1989; Pogribny et al. 2004; Pogribny et al. 2005; Kaup et al. 2006; Aypar et al. 2011; Goetz et al. 2011; Antwih et al. 2013; Kim et al. 2013; Lee et al. 2015), while others quantified DNA methylation in repetitive elements such as long and short interspersed nuclear elements (LINEs and SINEs) to estimate the potential risk of genomic instability after irradiation (Aypar et al. 2011; Lee et al. 2015). Only few studies screened the whole genome in an unbiased fashion (Kamstra et al. 2018; Thaulow et al. 2020; Laanen et al. 2021). Impey et al. reported on the murine methylome after exposure to protons (Impey et al. 2016), whereas Maierhofer et al. examined the radiation-induced response of human fibroblasts on a global scale (Maierhofer et al. 2017). A common trend that can be inferred from many of these studies is that radiation frequently induces focal hypermethylation of CpG islands, while DNA gets hypomethylated at the global scale (Miousse et al. 2017). This observation is reminiscent of DNA methylation changes observed during tumorigenesis and has fueled speculation as to what extent irradiation-induced silencing of genes by hypermethylation of CpG islands or genomic instability triggered by global loss of DNA methylation may contribute to tumorigenesis (Koturbash et al. 2005).

However, most studies mentioned above investigated DNA methylation changes in response to doses above 100 mSv. To the best of our knowledge, so far, only one study examined DNA methylation changes in mice after singular low X-ray exposure (10 mGy) (Newman et al. 2014). In contrast, no data at all are available on the impact of CT on DNA methylation. Therefore, epigenotoxic effects after low dose exposures such as CT scans remain elusive.

In this study, we performed a genome wide screen for changes of DNA methylation in response to low-dose CT exposure. A modern dual source CT scanner was employed to expose peripheral blood ex vivo following single-energy CT and DECT protocols, which enabled to check the existence of any CT-induced changes of DNA methylation and to evaluate the possible impact of applied photon spectra on this epigenetic mark. The similarities of the experimental design with a previous study on gene expression after CT exposure further allowed a direct comparison of both methods as to the usability as biological sensors of low-dose exposure.

Materials and methods

Ex vivo CT exposure and dose estimates

Ethical approval was granted by the ethics committee of the Medical Association of Rhineland Palatinate, Germany (reference number: 837.084.17(10918)). DNA for methylation analysis was isolated from aliquots of peripheral blood that have been taken in the course of a previous gene expression study. Thus, the experimental setup for ex vivo CT irradiation and dose estimation are identical to what has been described in detail there (Kaatsch et al. 2021). In brief, peripheral blood samples, drawn from three healthy individuals in duplicates, were placed inside a lumbar spine phantom and were irradiated in a modern dual source CT scanner (SOMATOM Force, Siemens Medical Solutions, Forchheim, Germany) applying single-energy CT (80 kV and 150 kV) and DECT (80 kV/Sn150 kV) protocols. Manual adjustment of tube voltage and tube current was performed to ensure comparable dose values (tube current: 950 mA 80 kV SECT, 160 mA 150 kV SECT, 440 mA 80 kV DECT and 220 mA 150 kV DECT). CT Dose Index (CTDIvol) was 17.86–18.26 mGy and dose length product (DLP) was 606.59–613.76 mGy*cm. The Monte Carlo model-based calculation of effective dose resulted in 7.0 ± 0.08 mSv (ImpactDose, CT Imaging GmbH, Erlangen, Germany).

Sample preparation and targeted methylation sequencing

After incubation for 6 h at 37 °C, blood samples were stored at −20 °C. Genomic DNA was extracted from thawed blood samples using the Qiagen Blood and Cell Culture DNA Mini Kit in accordance to the manufacturer’s recommendations (Qiagen, Venlo, NL, protocol version 06/15). The quality and quantity of extracted DNA were measured spectrophotometrically (NanoDrop, PeqLab Biotechnology, Erlangen, Germany) and fluorometrically (Qubit 3.0 Flourometer, Thermo Scientific, Waltham, MA). Targeted methylation sequencing (Methyl-Seq) was performed using the TruSeq Methyl Capture EPIC Library Prep Kit (Illumina Inc., San Diego, CA) on a total of 500 ng of input DNA in accordance to the manufacturer’s protocol (version #1000000001643v01, May 2017). In brief, fragmented genomic DNA was ligated to indexed adapters and genomic regions of interest were captured using a streptavidin-based pull down of biotin-labeled baits. Enriched target sequences were bisulfite treated, which converts unmethylated cytosine to thymine but leaves methylcytosine unmodified. Samples were sequenced on the Illumina NextSeq500 sequencing platform (2 × 150 bp reads; high output reagent kit; Illumina Inc., San Diego, CA).

Methyl-Seq data processing

After conversion of Illuminas proprietary .bcl files to FASTQ format, FASTQ files were trimmed with TrimGalore (Krueger 2015). Quality of bisulfite sequencing was assessed with FastQC (v0.11.9; Andrews 2010). Both, alignment to the reference genome (hg19) and methylation extraction were performed by means of bismark (v0.23.0; Krueger and Andrews 2011). Conversion efficiency of cytosines in the non-CpG context reflects the efficiency of bisulfite treatment. It was 99.8% on average, meeting the expectations for a successful bisulfite treatment. Resulting .cov files were imported into R (v4.05; R Core Team 2013) and differential methylation profiles were determined using dmrseq (v0.99.0; Korthauer et al. 2019). In order to exclude any effects of sequencing depth on the outcome, all analyses were done both with original, full-sized FASTQ files and with FASTQ files that were reduced to the same size by subsampling employing seqtk (Li 2013). For dmrseq, CpG count matrices were filtered for 1× coverage of CpG loci across all samples. DMRs were identified by setting a cutoff of 0.05 for single CpG coefficient, the length of smoothing spans to 1000, the minimum number of CpGs in a smoothing span window to 10, the maximum number of basepairs in between neighboring CpGs to 10,000, the maximum number of basepairs in between neighboring CpGs within the same DMR to 5000 referring to the author’s recommendation for targeted Methyl-Seq. For verification purposes, the bioinformatic analysis was repeated with MethylSig (v1.4.0; Park et al. 2014), an alternative R package for DNA methylation analysis. For MethylSig, CpG loci with a coverage less than 5 and more than 500 were removed and tiling of the genome was applied with a tile size of 500. Prior to testing for differential methylation with the MethylSig test, loci with group coverage less than 2× were removed. Regions that scored an FDR <0.1 were considered significant for both packages. Unsupervised hierarchical clustering analysis (Euclidean distance, linkage rule: ward.D2) was performed by standard R function hclust() based on prefiltered methylation estimates obtained by the bsseq package (v1.29.0; Hansen et al. 2012).

Results

In order to reveal any bias caused by different sequencing depths between samples, downstream analysis was performed twice: once including all successfully mapped reads and once after equalizing read numbers of all samples to 32.99 million paired-end reads, which was the lowest number observed among samples. Both approaches yielded nearly identical results. The data presented in Figure 1, Supplementary Figure 1 and Table 1 refer to the analysis based on equalized read numbers.

Figure 1. Unsupervised hierarchical clustering (Euclidean distance, Ward.D2 algorithm) using prefiltered DNA methylation estimates of all 24 Methyl-Seq experiments. Sample names include treatment (80 kV, 150 kV, DE and NULL) and healthy individuals (G1, G2 and G3). Healthy individuals are color coded in red (G1), green (G2) and blue (G3). The clustering analysis results in individual-specific clusters and a random distribution of the samples within each cluster without clearly distinguishable influence of the treatment.

Table 1. Inter-individual differences of DNA methylation and gene expression.

	Comparison of non-irradiated control samples
	G1 vs. G2	G1 vs. G3	G2 vs. G3
Number of DMRs	1224	4550	1163
Number of DMRs overlapping Ensembl genes	1021	3316	938
Differentially expressed genes	1348	2341	1678
Overlap among DEGs and DMRs with Ensembl gene mapping	67	438	88

Numbers of differentially methylated regions (DMRs), DMRs overlapping known Ensembl genes, differentially expressed genes (DEGs) from a previous study (Kaatsch et al. 2021) comparing non-irradiated control samples of each healthy individual and the overlap of differentially expressed genes with DMRs comprising Ensembl genes. Healthy individuals are indicated by G1, G2 and G3.

In the initial exploratory analysis, similarities between samples were evaluated by unsupervised hierarchical clustering based on DNA methylation patterns. Figure 1 depicts the result of unsupervised clustering employing the Ward.D2 algorithm after filtering all genomic positions with missing values in any of the samples. The result of hierarchical clustering without this filtering step is provided in Supplementary Figure 1. Both dendrograms revealed that clusters’ organization was mainly determined by inter-individual differences and not by treatment, suggesting very few, if any, CT-associated changes of DNA methylation (Figure 1). In line with this observation, the search for differentially methylated regions (DMRs) by means of dmrseq with commonly applied cutoffs (FDR <0.1) delivered 1163–4550 significant DMRs for the inter-individual comparison (Table 1), but failed to identify a single DMR in the comparison of irradiated and non-irradiated samples. This trend was confirmed using MethylSig starting from the non-equalized data. DMRs identified by MethylSig based on the input files with downsampled read numbers did not reach statistical thresholds after correction for multiple testing.

The availability of RNASeq gene expression data derived from the same samples (Kaatsch et al. 2021) enabled us to correlate DMRs and mRNA transcript abundance. For this comparison, we focused on DMRs overlapping known Ensembl genes (Table 1). As can be inferred from Table 1, only a small fraction of DMRs identified by comparison of non-irradiated samples map to genes that were found differentially expressed in the corresponding pairwise comparison of RNASeq data. No correlation of DNA methylation and gene expression was found when scrutinizing the DNA methylation characteristics of irradiated samples at AEN, BAX, DDB2, EDA2R and FDXR. These five genes were recurrently upregulated in all samples upon CT exposure with all photon spectra (Kaatsch et al. 2021). However, neither statistical analysis using dmrseq and MethylSig nor visual inspection by means of the integrative genomics browser (IGV) (Robinson et al. 2011) at the level of single sequencing reads provided evidence for CT-induced changes of DNA methylation at these sites.

Discussion

DNA methylation is a central regulator of gene expression with great impact on genome integrity. Despite its versatile role in many biological processes, to the best of our knowledge, the effect of low-dose exposure during CT on DNA methylation profiles has not yet been investigated.

In line with our previous studies on gene expression changes after CT exposure, performed with identical experimental settings and healthy individuals, inter-individual differences dominated the outcome of unsupervised hierarchical clustering based on sample to sample differences of DNA methylation patterns (Kaatsch et al. 2021). Accordingly, we have identified several genomic regions differentially methylated between individuals. However, unlike in previous studies by our group and others, where it was possible to discriminate irradiated from non-irradiated samples based on upregulation of genes and higher numbers of DNA double-strand breaks (Kaatsch et al. 2020, 2021), respectively, no single DMR was detected that could serve as marker for the separation of irradiated and non-irradiated samples. This applied to all tube voltages tested and corroborates previous long-term observations in mice, which indicated no DNA methylation changes in murine repetitive elements after 10 mGy (Newman et al. 2014). The lack of detectable DNA methylation differences suggests that low-dose exposure in the course of CT scans may not significantly contribute to the cellular response to CT exposure at the investigated point of time. However, this conclusion must be stated with reservations. Given the fact that the analysis platform records DNA methylation values averaged over the whole cell population, alterations only occurring in a small fraction of cells may remain undetected. The contribution of DNA methylation changes in single cells might be limited with respect to short-term effects but may gain relevance in the context of clonal expansion and long-term effects. Noteworthy, this technological limitation also applies to all other methods that do not assess cellular responses to irradiation at the single cell level.

With targeted high-throughput bisulfite sequencing, we have chosen a highly sensitive analysis method that facilitates comprehensive detection of DNA methylation with particular strength in gene regulatory regions. Nevertheless, we failed to identify any differentially methylated genomic interval in the same samples, where gene expression analysis showed CT-associated upregulation of AEN, BAX, DDB2, EDA2R and FDXR. This was unexpected given the known mutual interplay of DNA methylation and gene expression, suggesting that gene expression analysis may represent a more sensitive tool for detecting early effects of low-dose exposures such as CT scans. Based on the current data set, it can only be speculated on the underlying causes. For example, posttranscriptional modifications could exert influence on the stability of mRNA molecules resulting in higher transcript abundance without the need of increased transcription rates (Schaefke et al. 2018; Boo and Kim 2020). Alternatively, the response to low dose exposure could be heterogeneous across the irradiated cell population. Possible DNA methylation changes in only a few cells might escape detection, whereas each of these few cells could give rise to numerous transcripts that accumulate beyond the sensitivity thresholds of gene expression analysis. Further studies at the single cell level are needed to shed light on the possible heterogeneity of the cellular response to low-dose irradiation. Additionally, longer observation periods could provide clues on indirect effects impacting DNA methylation as well as effects that surface later by clonal expansion of affected cells.

Disclosure statement

The authors declare that there is no conflict of interest.

Data availability statement

The data that support the findings of this study are available on request from the corresponding author (RU).

Additional information

Notes on contributors

Benjamin Valentin Becker

Benjamin Valentin Becker, MD, is a Consultant Radiologist at the Bundeswehr Central Hospital Koblenz, Germany.

Hanns Leonhard Kaatsch

Hanns Leonhard Kaatsch, MD, is a resident in Radiology and a Post-Doctoral Researcher of Radiobiology at the Bundeswehr Institute of Radiobiology, Munich, Germany.

Kai Nestler

Kai Nestler, MD, is a resident in Radiology at the Bundeswehr Central Hospital Koblenz, Germany.

Julia Jakobi

Julia Jakobi is a technical laboratory assistant at the Bundeswehr Institute of Radiobiology, Munich, Germany.

Barbara Schäfer

Barbara Schäfer is a technical laboratory assistant at the Bundeswehr Institute of Radiobiology, Munich, Germany.

Thomas Hantke

Thomas Hantke is a technical laboratory assistant at the Bundeswehr Institute of Radiobiology, Munich, Germany.

Marc A. Brockmann

Marc A. Brockmann, MD, MSc, is a Full Professor of Neuroradiology and Chair of Department of Neuroradiology at the University Medical Center Mainz, Germany.

Stephan Waldeck

Stephan Waldeck, MD, is a Consultant Radiologist and Neuroradiologist and Head of Department of Radiology and Neuroradiology at the Bundeswehr Central Hospital, Koblenz, Germany.

Matthias Port

Matthias Port, MD, is a Professor of Radiobiology and Internal Medicine and Head of the Bundeswehr Institute of Radiobiology, Munich, Germany.

Reinhard Ullmann

Reinhard Ullmann, PhD, is a biologist and senior scientist at the Bundeswehr Institute of Radiobiology, Munich, Germany.