A molecule called DNA encodes genetic information inside our cells. Random changes to the DNA sequence, known as mutations, can occur in any cell. Most mutations are harmless, but some can lead to disease – most prominently cancer. Like how car accidents can happen more often on some roads than others, mutations are more frequent in some parts of the DNA. Cytosine, one of the four letters of the genetic code, usually accumulates more mutations than the other three letters.

Cytosine can be decorated with distinct ‘marks’ to form either methyl-cytosine or hydroxymethyl-cytosine. Methyl-cytosine is known to mutate relatively easily, and is the most common type of mutation observed in most cancers. However, little was known about how easily hydroxymethyl-cytosine mutates.

Modifications of cytosine are distributed differently in cells from different tissues. To test whether hydroxymethyl-cytosine mutates more or less often than methyl-cytosine in human cells, Tomkova et al. used the cytosine mutations measured in human brain, kidney and blood cancer samples. Comparing these mutations to maps of cytosine modifications from healthy tissues of the same type revealed that in all tissues, hydroxymethyl-cytosine appears to mutate less often than methyl-cytosine.

There are several possible explanations for the difference in mutation frequency between methyl-cytosine and hydroxymethyl-cytosine. Tomkova et al. plan to investigate these possibilities further in an effort to fully understand the underlying mechanisms that drive cytosine to mutate.

Cancer genomics projects have revealed that the distribution of somatic mutations across the genome is not uniform (Lawrence et al., 2013). Apart from positive and negative selective pressure, a number of factors can influence regional mutation frequencies, such as chromatin organisation (Schuster-Böckler and Lehner, 2012), replication timing (Koren et al., 2012), metabolic load (Ames et al., 1993) and exposure to different mutagens (Poon et al., 2013). Furthermore, highly transcribed regions generally exhibit lower mutation frequencies due to the influence of transcription-coupled repair (Lawrence et al., 2013). In addition to the regional distribution of mutations, the local nucleotide contexts and mutation types (referred to as mutational signatures) have been investigated extensively since they provide critical clues about the mechanism of mutagenesis. For example, consensus motifs for cytidine deaminases (such as APOBEC and AID) were found enriched at mutational hot-spots, suggesting that activity of these enzymes could be the potential cause of those mutations (Nik-Zainal et al., 2012; Taylor et al., 2013). The most frequent mutational signature found in the majority of cancers is C to T transition in a CpG dinucleotide context (CpG>T) (Alexandrov et al., 2013; Lawrence et al., 2013). This relates to the fact that cytosines in CpG dinucleotides are frequently methylated to form 5-methylcytosine (5mC). The rate of spontaneous deamination of 5mC into T is four fold higher than the rate of deamination of C into U (Lindahl and Nyberg, 1974). In the germline of vertebrates, this likely facilitated a general depletion of CpGs outside of CpG islands which are largely unmethylated.

The genomes of all examined vertebrate species feature DNA methylation, and loss of DNA methylation is incompatible with normal development in mice (Li et al., 1992; Okano et al., 1999). DNA methylation plays a role in gene expression, most notably by repressing one allele of imprinted genes. Moreover, it is involved in maintenance of genome stability, alternative splicing, X chromosome inactivation and suppression of retrotransposons (Klose and Bird, 2006; Jones, 2012). In 2009, 5-hydroxymethylcytosine (5hmC) was indisputably shown to exist in DNA of brain and other tissues (Kriaucionis and Heintz, 2009). It was concurrently shown that ten-eleven translocation (TET) enzymes are able to convert 5mC into 5hmC (Tahiliani et al., 2009). Unlike 5mC, which is observed at similar levels in many cell types, the abundance of 5hmC varies widely. 5hmC was observed to be particularly enriched in brain cells (Kriaucionis and Heintz, 2009; Lister et al., 2013) and detectable in embryonic stem cells and all examined tissues (Tahiliani et al., 2009; Globisch et al., 2010; Szwagierczak et al., 2010; Wu and Zhang, 2011). 5hmC and higher oxidised states of methyl-cytosine have been proposed to play a role in de-methylation via ineffective re-methylation after replication or directly by thymine DNA glycosylase (TDG) (Tahiliani et al., 2009; He et al., 2011; Maiti and Drohat, 2011; Shen et al., 2013; Hu et al., 2014). In addition to demethylation, 5hmC has been implicated in transcriptional regulation, and a number of DNA binding proteins recognising 5hmC have been identified (Mellén et al., 2012; Spruijt et al., 2013; Takai et al., 2014). 5hmC is found depleted in primary tumours and TET2 is frequently mutated in myelodysplastic syndrome, acute myelogenous leukaemia and T-cell lymphoma, indicating that 5hmC plays a role in carcinogenesis. However, the molecular mechanism by which 5hmC affects carcinogenesis is poorly understood (Rasmussen and Helin, 2016).

5hmC is an important intermediate during demethylation in zygotes and ES cells (Tahiliani et al., 2009; Inoue and Zhang, 2011; Wossidlo et al., 2011), but the vast majority of 5hmC is found as a stable, long-lived modification in adult mouse tissue that undergoes little cell division (Bachman et al., 2014; Brazauskas and Kriaucionis, 2014). Thus, we hypothesised that – similar to 5mC – long-lived 5hmC could have a substantial influence on the mutability of DNA. Little is known about the mutational properties of 5hmC, in part because until recently there has been a lack of information on the precise location of 5hmC in the genome. With the development of techniques for single-nucleotide resolution mapping of 5mC and 5hmC (Yu et al., 2012; Booth et al., 2014), it is now possible to differentiate mutation rates at 5mC and 5hmC sites. Recently, Supek et al. (Supek et al., 2014) reported elevated C>G transversion rates at 5hmC sites, using 5hmC maps from human and mouse embryonic stem cells. However, these findings are limited by the fact that embryonic stem cells differ substantially from the somatic tissues in which mutations were observed (Schultz et al., 2015).

A large proportion of mutations observed in any cancer genome originate in its pre-cancerous cell of origin (Nik-Zainal et al., 2012; Stephens et al., 2012; Tomasetti et al., 2013; Wu et al., 2015) and will have been influenced by its epigenetic landscape. The publication of single-base resolution maps of 5mC and 5hmC occupancy in samples of human brain, kidney and blood (Wen et al., 2014; Chen et al., 2015; Pacis et al., 2015) now enables us to interrogate the tissue-specific effect of cytosine modifications on somatic mutation rates in unprecedented detail.

Since 5hmC has been shown to be most abundant in human brain (Li and Liu, 2011; Nestor et al., 2012), we have initially focussed on assessing the relationship between mutability and DNA modifications in brain cancers. Based on a DNA sequencing data from five brain cancer types encompassing 665 patients, we show that the dominant mutational signature in brain cancers is CpG>T, which is modulated by the modification state of cytosine. Strikingly, the CpG>T mutation frequency of 5-hydroxymethylcytosine is reduced nearly two-fold compared to the methylated state. We find that the ratio of 5hmC to 5mC in 100 kb genomic intervals correlates with CpG>T mutation frequency even after accounting for confounding factors like gene density or CpG islands. When we expand our analysis to include mutations and 5hmC maps from kidney and myeloid lineage of blood cells, we observe a clear tissue-specific effect of 5hmC on mutagenicity. Finally, we measured 5mC and 5hmC levels using methodology of high accuracy in eight different human tissue types and show that reduced 5hmC levels associate with an increased proportion of CpG>T mutations in cancers of the corresponding tissue. Together, our findings suggest that hydroxymethylation has a significant influence on the likelihood of mutations at CpG sites across many human tissue types.

We compiled base-resolution maps of 5mC and 5hmC frequency in brain, kidney and myeloid cells from publicly available sources (Wen et al., 2014; Chen et al., 2015; Pacis et al., 2015). All three data sets are based on bisulfite (BS) and 'Tet-assisted bisulfite' (TAB) sequencing, respectively. BS-Seq detects any modified cytosine (i.e., does not distinguish 5mC and 5hmC) whereas TAB-Seq specifically detects 5hmC. The combination of the two methods allows an estimation of the levels of both 5mC and 5hmC for all sufficiently covered cytosines. As 5hmC predominantly occurs in a CpG context, we focussed the analysis on CpG sites. Sequencing reads come from heterogeneous populations of cells. Hence, a single locus usually cannot be assigned a single state (C, 5mC or 5hmC). Instead, we estimated the frequency of modification, hydroxymethylation and methylation per site using the percentage of BS-Seq reads that showed a modification (referred to as mod level), the percentage of TAB-Seq reads that showed hydroxymethylation (referred to as 5hmC level) and their difference (5mC level = mod level – 5hmC level), respectively.

5hmC sites in brain exhibit lower frequency of CpG>T mutations than 5mC sites

Since brain tissue has been shown to exhibit particularly high levels of 5hmC (Figure 1A), we first investigated the relationship between the regional distribution of 5hmC, 5mC and mutagenesis in brain tumours. We reasoned that this approach would provide the highest sensitivity to detect any correlation between 5hmC and mutation frequency.

Figure 1 with 3 supplements see all

Download asset Open asset

We analysed 344370 somatic single nucleotide variants (SNVs) from 665 samples derived from exome and whole genome sequencing of the following cancer types: Glioblastoma (GBM), Glioma low grade (GLG), Neuroblastoma (NRB), Medulloblastoma (MDB) and Pilocytic astrocytoma (PA) (Alexandrov et al., 2013). The dominant point mutation type in these cancers were C>T transitions in a CpG context (Figure 1B,C), similar to what was observed in combined analyses of all cancer types (Alexandrov et al., 2013).

Mutations and DNA modifications are not distributed uniformly along the chromosomes. Strikingly, 5hmC levels were visibly and significantly anti-correlated with the frequency of CpG>T (r=-0.71, chr3), while 5mC levels displayed a positive correlation (r=0.66, chr3, Figure 1D, Figure 1—figure supplements 1–2). This is not a simple consequence of the uneven distribution of genes, exons, CpG islands or levels of gene expression (Figure 1—figure supplement 3 and additional analyses below).

Averaging over the entire genome, the frequency of C>T mutations differed substantially between 5mC_high (5mC_high: mod level > 10% and 5hmC level / mod level ≤ 0.3) and 5hmC_high (5hmC_high: mod level > 10% and 5hmC level / mod level ≥ 0.5) sites. The fraction of mutated 5hmC_high sites was significantly lower than the fraction of mutated 5mC_high sites (Figure 1E). The lower mutation frequency was consistently observed in data derived from both exome and whole genome sequencing projects (p<0.001, Wilcoxon signed-rank test). Moreover, all brain cancer types individually displayed a significant (28–53%, p<0.05 in all types) reduction of C>T mutations in 5hmC_high sites (Figure 2A,B).

Figure 2 with 1 supplement see all

Download asset Open asset

It has been shown that CpG>T mutations are one of the two mutational signatures correlating with age (Alexandrov et al., 2015), supporting the fact that these mutations were gathered during the entire lives of the patients, not only after the origin of cancer. Here we observed that this correlation is present in both methylated and hydroxymethylated sites (Figure 2C). Moreover, the slope for 5mC was steeper than for 5hmC, suggesting that even the mechanisms causing the difference of CpG>T mutability between 5mC and 5hmC were present in the pre-cancerous cell of origin.

We also compared the fraction of mutated 5mC_high and 5hmC_high sites for the other two possible types of mutations: C>A and C>G. As shown in Figure 1E, C>A or C>G transversions are an order of magnitude less frequent than C>T transitions in both 5mC and 5hmC sites. The relationship between C>A and C>G mutations and 5hmC varied between cancer types. In GBM and LGG the frequency of C>A mutations was significantly higher in 5mC_high compared to 5hmC_high sites, but in NRB, MDB and PA we detected no significant difference. The frequency of C>G mutations in 5mC_high sites differed significantly from 5hmC_high sites only in MDB, PA and GBM. In MDB and PA, 5hmC_high sites were slightly enriched for C>G mutations, whereas in GBM an enrichment was observed at 5mC_high sites. Since C>T transitions are the most common somatic mutation type in brain and the difference in C>T mutations between 5mC_high and 5hmC_high sites is more consistent among cancer types than in the C>A and C>G transversions, we focus mainly on C>T mutations in the remainder of this report.

We confirmed that C>T mutations are significantly depleted at 5hmC sites across a wide range of thresholds in definitions of 5mC_high and 5hmC_high (Figure 2—figure supplement 1A–F). In fact, more stringent definitions of 5hmC (e.g., 5hmC_high: 5hmC level / mod level ≥ 0.7) result in even greater differences (42–59%) in C>T mutation frequencies between 5mC_high and 5hmC_high sites (Figure 2—figure supplement 1G–I, Figure 3—figure supplement 1A–D), but these definitions would reduce the number of sites too much for our subsequent statistical analyses.

Reduced 5hmC mutability in brain is not accounted for by genomic regions or gene expression

We next examined whether the decreased frequency of C>T transitions at 5hmC vs. 5mC sites might be an indirect effect of 5hmC being associated with genomic regions of lower mutability. Levels of 5mC and 5hmC vary across genomic regions. For example, 5hmC density is elevated in highly expressed genes in brain (Mellén et al., 2012; Yu et al., 2012; Lister et al., 2013; Wen et al., 2014). The observed decrease in C>T mutation frequencies might therefore be attributable to higher gene expression, which would correlate with higher transcription coupled repair. We therefore performed the analysis described above separately for regions with high vs. low gene expression in human brain (see Materials and methods). There was a lower overall mutation frequency in highly expressed genes (Figure 3A–B), but both highly and lowly expressed genes exhibited significantly lower C>T transition rates at 5hmC sites compared to 5mC sites (Figure 3A–D). This suggests that the observed effect is independent of transcription and thus not a result of transcription coupled repair.

Figure 3 with 1 supplement see all

Download asset Open asset

Gene expression is only one of many possible region-related confounding factors. Hence, to correct for any regional variation, we performed the analysis on groups of sites generated by pairing the modified CpGs: each 5hmC site was paired with the nearest yet unpaired 5mC site from an equivalent genomic and sequence context (an approach adapted from Supek et al., 2014, see Materials and methods). Thereby we compared the mutation frequencies of two groups (one group comprising 5mC sites and one group comprising 5hmC sites) containing the same number of loci (6801374 cytosines in each group). As a result of this experimental setup, a substantial fraction of mutated 5mC sites were excluded, greatly reducing the statistical power of this 'paired' analysis. Nevertheless, the frequency of C>T mutations in 5hmC remained significantly lower than in 5mC in both exomes and genomes (Figure 3E–F), supporting that the difference between 5mC and 5hmC mutation frequency is not caused by regional differences.

To ensure that there is no confounding bias in the spatial distribution of mutations around 5mC or 5hmC sites, respectively, we plotted mutation frequencies in a 2 kb radius up and downstream of modified loci (Figure 3—figure supplement 1G, Materials and methods). The mutation frequency differed substantially in the aligned positions of DNA modifications but was indistinguishable in the surrounding area. In conclusion, regional mutation rate variability is unlikely to account for the difference in C>T mutational load in 5mC and 5hmC sites.

Relative 5hmC correlates with CpG>T mutation frequency

The 5mC and 5hmC frequency at each base reflect the prevalence of each modification within the sequenced cell population. We hypothesised that if 5hmC had a direct effect on C>T mutation likelihood, we would observe an increase in mutation frequency with decreasing 5hmC occupancy. To test this, we formally defined 5hmC_rel as the ratio of the proportion of reads supporting 5hmC, relative to the proportion of reads supporting any modification at a particular cytosine (5hmC_rel = 5hmC level / mod level). We then divided brain CpG sites into bins according to their 5hmC_rel level and computed the fraction of mutated sites in each bin (Figure 4A). We observed a clear linear relationship between 5hmC_rel values and C>T mutation frequencies. Notably, the correlation was present in all the tested brain cancer types in exome- and whole genome-sequenced samples. A regression slope test confirmed significance of this relationship in all the cancer types. To confirm that the results are not influenced by an uneven distribution of information across bins, we performed quantile binning so that each bin contains an approximately equal number of positions (see Materials and methods). The results of quantile bins were equivalent to evenly spaced bins (Figure 4—figure supplement 1H).

Figure 4 with 1 supplement see all

Download asset Open asset

For comparison, we also evaluated the relationship between 5hmC_rel and the frequency of C>A and C>G mutations (Figure 4A). Consistent with our previous results, an increase in 5hmC_rel is associated with an increase in C>G mutations in whole genomes (from MDB and PA samples), but the relationship in other cancer types shows no significant trend. C>A mutations decrease with 5hmC_rel levels in GBM but exhibit no significant signal in the remaining tumour types.

This result supports the conclusion that the decrease in C>T mutation frequency at 5hmC sites is not an artefact of our chosen definition of 5mC or 5hmC. Even more importantly, it supports the notion that this decrease is directly caused by the properties of these DNA modifications.

5hmC is a predictor of CpG>T mutation frequency across the genome

To examine the exclusive impact of DNA modifications on regional frequencies of mutations, we split the genome into 100 kb windows and fitted a generalised linear model to explain the observed per-window CpG>T mutation frequency from a combination of features including average 5mC and 5hmC levels, 5hmC_rel, gene density, CpG island density amongst others. Only whole genome sequencing data were used for this analysis. To compare the resulting models, we calculated their respective 'explained deviance' D², a generalisation of explained variance that is more appropriate for comparing generalised linear models (see Materials and methods).

The best individual predictor of CpG>T mutation frequency was 5hmC_rel (D² = 0.11), outperforming all other features (Figure 5A). Interestingly, the sum of 5mC and 5hmC levels ('mod') performed worst, suggesting that bisulfite sequencing measurements alone are a poor predictor of mutagenicity. When combining all 11 features into one model, the total explained deviance for 100 kb windows was 16%.

Figure 5 with 3 supplements see all

Download asset Open asset

Varying the chosen window size (3 kbp – 3 Mbp; Figure 5B, Figure 5—figure supplement 1A–C) does not substantially change the comparison of the predictive power of the respective features. In all cases, 5mC and 5hmC_rel were the two best predictors, with 5hmC_rel performing slightly better with smaller windows. However, the total explained deviance increased with window size, reaching values as high as 45% for univariate models and 60% for models with all predictors. This led us to question whether the increasing predictive power of larger windows has a biological reason, or whether it is a consequence of the lower data density in small windows.

Since many smaller windows contain no observed mutations, low D² values could simply reflect a lack of data. To test this, we generated simulated mutations so that a 'perfect' predictor was linearly related to the mutation likelihood per window (see Materials and methods). We then measured the effect of window and sample size (number of patients) on the observed D², repeating the simulations 10 times. The resulting curves of the explained deviance resemble those measured in the real data (Figure 5—figure supplement 1D). Moreover, in the simulated data, higher numbers of patients lead to higher D² even for smaller window sizes, suggesting that lower D² values in smaller windows indeed are a consequence of lower data density.

Decreased CpG>T mutation frequency in 5hmC is not limited to brain tissue

Two recently published datasets allowed us to address the question of mutational properties of 5mC and 5hmC also in two other tissues: kidney (Chen et al., 2015) and blood (Pacis et al., 2015). For blood we used 174 sequencing samples from Acute Myeloid Leukaemia (AML) as the cancer type closest to the blood dendritic cells in which the BS-Seq and TAB-Seq measurements were performed. For kidney we combined 585 samples from four distinct sequencing projects, covering Kidney Clear Cell, Kidney Papillary and Kidney Chromophobe carcinomas.

Matching our findings in brain, 5hmC sites were mutated significantly less frequently than 5mC sites in both tissue types (Figure 6B), irrespective of whether genome or exome sequencing data were used. Moreover, a similar difference was present in all available replicates of the BS-Seq and TAB-Seq measurements (6 for blood, 2 for kidney, Figure 6—figure supplement 1A).

Figure 6 with 3 supplements see all

Download asset Open asset

Genomic distribution of 5hmC differs substantially between the three tissue types (Figure 6—figure supplement 2). Consequently, we expected the association between mutation frequency and 5hmC to be highest when mutation and modification data are derived from matching tissue types. To test this hypothesis, we used a GLM on genomic windows of 100 kbp to predict CpG>T mutation rate from a combination of 5hmC_rel maps of all three tissues. Strikingly, for each cancer type, the best predictor came from the same tissue type (Figure 6A), suggesting that tissue differences in 5hmC are reflected in the CpG>T mutation landscape. The same results were obtained in all available replicates of the 5hmC_rel maps (Figure 6—figure supplement 1B). Finally, we added a 5hmC_rel map derived from embryonic stem cells (ESC) as an additional predictor, to compare our findings to previously reported results (Supek et al., 2014). As we anticipated, the ESC-derived 5hmC levels have substantially lower predictive power on CpG>T mutation rate than any of the tissue-derived maps.

While base-resolution maps of 5hmC for human tissue are still rare, there is a wide range of BS-Seq data sets available in public databases. Given our findings thus far, we predicted that tissues with high levels of 5hmC relative to 5mC would exhibit fewer CpG>T mutations in modified sites than tissues with low total 5hmC. To test this hypothesis, we measured total levels of 5mC and 5hmC using High Pressure Liquid Chromatography (HPLC-UV) in DNA of eight human tissue types for which BS-Seq maps are publicly available (Figure 6—figure supplement 3). In order to account for unrelated cancer-type specific differences in CpG>T mutability, we normalised the mutation frequency in modified sites by the mutation frequency in unmodified sites. The analysis of association between genomic 5hmC and enrichment of CpG>T mutations revealed a strong negative correlation (Figure 6C) in all tissue types except lung. Interestingly, this difference seems to stem from smoking-related effects. Lung cancer mutation data from heavy smokers revealed a markedly lower frequency of CpG>T mutations in modified sites, relative to other mutations. It has been reported that the typical C>A mutational signature associated with smoking was found significantly enriched in CpGs outside CpG islands, suggesting that it preferentially occurs at modified CpG sites (Pleasance et al., 2010). Accordingly, our data indicate that CpG>T mutations might also be differentially affected by smoking-related mutagens.

Here we have established a link between the landscape of DNA modifications and the mutational profile of somatic human cells. Our measurements indicate that 5hmCs carry between 28 and 53% fewer mutations than methylated cytosines in brain. This results in a mutational load at 5hmC sites that is comparable to that of unmodified cytosines in CpG dinucleotides. This effect is not only observable in brain, but also in kidney cancers and myeloid leukaemias. The relationship between 5hmC and CpG>T mutation rate can be detected at the scale of the exome as well as genome-wide and is independent of other region-specific influences on mutation frequency. We show that the relative impact of hydroxymethylation on mutagenesis decreases proportionally to the level of 5hmC in the tissue, suggesting that it represents a general property of this DNA modification.

Two possible scenarios could explain the striking difference in mutability between 5mC and 5hmC. Firstly, spontaneous and enzymatic deamination reactions of 5hmC could be less favourable than 5mC. As a consequence, fewer new mutation events would be expected at 5hmC sites. Indeed, cytosine deaminases (namely, AID and APOBEC1-3) have 4.4–38x lower activity on sites with 5hmC compared to 5mC, supporting this possibility (Nabel et al., 2012; Rangam et al., 2012). Secondly, deamination of 5mC produces thymine whereas 5hmC deaminates to 5-hydroxymethyluracil (5hmU). This atypical base in DNA could be more efficiently recognised and replaced by the DNA glycosylases initiating base-excision repair (BER). Determining the relative contribution of DNA glycosylases to the lower mutation rate would be challenging, since some of these enzymes recognise several types of mismatches. TDG and MBD4 excise both T and 5hmU when mis-paired with G (Hardeland et al., 2003; Cortellino et al., 2011; Guo et al., 2011; Hashimoto et al., 2012; Moréra et al., 2012), whereas SMUG1 does not repair T:G but has a robust activity for 5hmU:G (Nilsen et al., 2001; Kemmerich et al., 2012). Therefore, there might be more efficient repair of 5hmU in the genome. Further genome sequencing efforts might identify patients with rare inactivating mutations in BER and/or mismatch-repair pathways that could be valuable for future investigations of the relationship between DNA repair and cytosine mutability.

It has previously been suggested that 5hmC levels increase the frequency of C>G mutations (Supek et al., 2014). As part of this analysis, only a very small (albeit statistically significant) decrease of C>T mutations in 5hmC sites in both SNPs and cancer SNVs was observed. There are two factors that could explain why we observe very different effects sizes for C>T and C>G mutations in 5hmC sites. Firstly, Supek et al. consider all sites with as little as one 5hmC read to be hydroxymethylated, whereas we require the level of 5hmC to exceed 5mC. In fact, when examining the effect of variation in these thresholds (Figure 2—figure supplement 1A–F), we noticed that the results for C>G fluctuate substantially across the range of tested cut-off values (see also Figure 3—figure supplement 1E–F). Secondly, we present evidence that tissue-specific changes in 5hmC patterns have great influence on the extent of correlation between 5hmC and mutability (Figure 6B). Specifically, 5hmC genomic localisation in embryonic stem cells was a poor predictor of CpG>T mutations in brain, kidney and blood, compared to the respective tissue-specific 5hmC patterns.

The best predictor of CpG>T mutations in any of the three tested tissues was the 5hmC_rel map from the corresponding anatomical site. This provides evidence that the slow accumulation of CpG>T mutations in the pre-cancerous tissue was strongly influenced by the DNA modification landscape. However, any bulk tissue sample encompasses a mixture of different cell types. Mounting evidence suggests that solid tumours originate from a defined subset of cells within any one tissue type. For example, glioblastomas were proposed to originate from stem or progenitor cell types enriched in the subventricular zone, while medulloblastomas have mixed cells of origin (Visvader, 2011). Those cell types are of low abundance in normal tissue biopsies. The fact that we observe a clear inverse relationship between CpG>T mutations and the location of 5hmC in multiple tissue types suggests that the DNA modification landscape in cancer-progenitor cells is sufficiently similar to the tissue average to be informative about the mutation frequencies in cancer.

Under this assumption we predict that the impact of DNA modifications on the frequency of CpG>T mutations is likely to be bigger than measured here, since the terminally differentiated cells that make up the bulk of the tissue may have diverged further from cancer-progenitor cells. Advancements in the identification of cancer origins and isolation of single cells, combined with single-cell bisulfite sequencing, will enable an improved assessment of the impact of DNA modifications on mutability.

The strong correlation between relative 5hmC levels in a tissue and the mutability of modified cytosine also points towards a shared underlying mutagenic process. The notable deviation of smoking-induced lung-cancers supports this hypothesis. We speculate that a yet undefined smoking-induced mutagenic mechanism preferentially affects unmethylated CpG sites. More experimental work will be needed to elucidate the biochemical causes for this phenomenon. In the future, the linear relationship between 5hmC levels and CpG>T mutation rate could thus be used to identify other environmental mutagens with a differential effect on modified cytosines.

1. 1. Alexandrov LB
   2. Nik-Zainal S
   3. Wedge DC
   4. Aparicio SA
   5. Behjati S
   6. Biankin AV
   7. Bignell GR
   8. Bolli N
   9. Borg A
   10. Børresen-Dale AL
   11. Boyault S
   12. Burkhardt B
   13. Butler AP
   14. Caldas C
   15. Davies HR
   16. Desmedt C
   17. Eils R
   18. Eyfjörd JE
   19. Foekens JA
   20. Greaves M
   21. Hosoda F
   22. Hutter B
   23. Ilicic T
   24. Imbeaud S
   25. Imielinski M
   26. Imielinsk M
   27. Jäger N
   28. Jones DT
   29. Jones D
   30. Knappskog S
   31. Kool M
   32. Lakhani SR
   33. López-Otín C
   34. Martin S
   35. Munshi NC
   36. Nakamura H
   37. Northcott PA
   38. Pajic M
   39. Papaemmanuil E
   40. Paradiso A
   41. Pearson JV
   42. Puente XS
   43. Raine K
   44. Ramakrishna M
   45. Richardson AL
   46. Richter J
   47. Rosenstiel P
   48. Schlesner M
   49. Schumacher TN
   50. Span PN
   51. Teague JW
   52. Totoki Y
   53. Tutt AN
   54. Valdés-Mas R
   55. van Buuren MM
   56. van 't Veer L
   57. Vincent-Salomon A
   58. Waddell N
   59. Yates LR
   60. Zucman-Rossi J
   61. Futreal PA
   62. McDermott U
   63. Lichter P
   64. Meyerson M
   65. Grimmond SM
   66. Siebert R
   67. Campo E
   68. Shibata T
   69. Pfister SM
   70. Campbell PJ
   71. Stratton MR
   72. Stratton M

   (2013) Signatures of mutational processes in human cancer

   Publicly available from the authors' website.

   ftp://ftp.sanger.ac.uk/pub/cancer/AlexandrovEtAl
2. 1. Wen L
   2. Li X
   3. Yan L
   4. Tan Y
   5. Li R
   6. Zhao Y
   7. Wang Y
   8. Xie J
   9. Zhang Y
   10. Song C
   11. Yu M
   12. Liu X
   13. Zhu P
   14. Hou Y
   15. Guo H
   16. Wu X
   17. He C
   18. Tang F
   19. Qiao J

   (2014) Whole-genome analysis of 5-hydroxymethylcytosines and 5-methylcytosines at base resolution in human brain

   Publicly available at the NCBI Gene Expression Omnibus (accession no: GSE46710).

   http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE46710
3. 1. The GTEx Consortium

   (2013) GTEx Analysis V4

   Publicly available at GTEx Portal (http://gtexportal.org).

   http://gtexportal.org/static/datasets/gtex_analysis_v4/rna_seq_data/GTEx_Analysis_V4_RNA-seq_RNA-SeQCv1.1.8_gene_rpkm.gct.gz
4. 1. Blueprint Epigenome Project Consortium

   (2012) Blueprint Epigenome

   Data available via application to the BLUEPRINT Data Access Committee (http://www.blueprint-epigenome.eu/index.cfm?p=B5E93EE0-09E2-5736-A708817C27EF2DB7).

   http://dcc.blueprint-epigenome.eu/#/experiments/ERX715127
5. 1. Internation Cancer Genome Consortium (ICGC)

   (2014) RENAL CELL CANCER - EU/FR

   Publicly available at the ICGC Data Portal (project: RECA-EU).

   https://dcc.icgc.org/projects/RECA-EU
6. 1. The Cancer Genome Atlas (TCGA)

   (2014) The Cancer Genome Atlas (TCGA)

   LAML WGS data available via CGHub.

   https://tcga-data.nci.nih.gov/tcga/
7. 1. Wang K
   2. Yuen ST
   3. Xu J
   4. Lee SP
   5. Yan HHN
   6. Shi ST
   7. Siu HC
   8. Deng S
   9. Chu KM
   10. Law S
   11. Chan KH
   12. Chan ASY
   13. Tsui WY
   14. Ho SL
   15. Chan AKW
   16. Man JLK
   17. Foglizzo V
   18. Ng MK
   19. Chan AS
   20. Ching YP
   21. Cheng GHW
   22. Xie T
   23. Fernandez J
   24. Li VSW
   25. Clevers H
   26. Rejto PA
   27. Mao M
   28. Leung SY

   (2014) Whole-genome sequencing and comprehensive molecular profiling identify new driver mutations in gastric cancer

   Publicly available at SY Leung's Laboratory website (http://web.hku.hk/~suetyi/).

   http://www.nature.com/ng/journal/v46/n6/full/ng.2983.html#supplementary-information
8. 1. Chen K
   2. Zhang J
   3. Guo Z
   4. Ma Q
   5. Xu Z
   6. Zhou Y
   7. Li Z
   8. Liu Y
   9. Ye X
   10. Li X
   11. Yuan B
   12. Ke Y
   13. He C
   14. Zhou L
   15. Liu J
   16. Ci W.

   (2015) Loss of 5-hydroxymethylcytosine is linked to gene body hypermethylation in kidney cancer

   Publicly available at the NCBI Gene Expression Omnibus (accession no: GSE63183).

   http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE63183
9. 1. Roadmap Epigenomics Consortium

   (2013) Roadmap Epigenome (Breast)

   Publicly available at the NCBI Gene Expression Omnibus (accession no: GSM1127125).

   http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM1127125
10. 1. Roadmap Epigenomics Consortium

    (2013) Roadmap Epigenome (Pancreas)

    Publicly available at the NCBI Gene Expression Omnibus (accession no: GSM983651).

    http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM983651
11. 1. Roadmap Epigenomics Consortium

    (2013) Roadmap Epigenome (Lung)

    Publicly available at the NCBI Gene Expression Omnibus (accession no: GSM983647).

    http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM983647
12. 1. Roadmap Epigenomics Consortium

    (2012) Roadmap Epigenome (Liver)

    Publicly available at NCBI Gene Expression Omnibus (accession no: GSM916049).

    http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM916049
13. 1. Roadmap Epigenomics Consortium

    (2013) Roadmap Epigenome (Stomach)

    Publicly available at NCBI Gene Expression Omnibus (accession no: GSM1010984).

    http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSM1010984
14. 1. Pacis A
    2. Tailleux L
    3. Morin AM
    4. Lambourne J
    5. Maclsaac JL
    6. Yotova V
    7. Dumaine A
    8. Danckaert A
    9. Luca F
    10. Grenier JC
    11. Hansen KD
    12. Gicquel B
    13. Yu M
    14. Pai A
    15. He C
    16. Tung J
    17. Pastinen T
    18. Kobor MS
    19. Pique-Regi R
    20. Gilad Y
    21. Barreiro LB

    (2015) Bacterial infection remodels the DNA methylation landscape of human dendritic cells (TAB-seq)

    Publicly available at NCBI Gene Expression Omnibus (accession no: GSE64181).

    http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE64181
15. 1. Pacis A
    2. Tailleux L
    3. Morin AM
    4. Lambourne J
    5. Maclsaac JL
    6. Yotova V
    7. Dumaine A
    8. Danckaert A
    9. Luca F
    10. Grenier JC
    11. Hansen KD
    12. Gicquel B
    13. Yu M
    14. Pai A
    15. He C
    16. Tung J
    17. Pastinen T
    18. Kobor MS
    19. Pique-Regi R
    20. Gilad Y
    21. Barreiro LB

    (2015) Bacterial infection remodels the DNA methylation landscape of human dendritic cells (BS-seq)

    Publicly available at NCBI Gene Expression Omnibus under (accession no: GSE64177).

    http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE64177

1. 1. Alexandrov LB
   2. Nik-Zainal S
   3. Wedge DC
   4. Aparicio SA
   5. Behjati S
   6. Biankin AV
   7. Bignell GR
   8. Bolli N
   9. Borg A
   10. Børresen-Dale AL
   11. Boyault S
   12. Burkhardt B
   13. Butler AP
   14. Caldas C
   15. Davies HR
   16. Desmedt C
   17. Eils R
   18. Eyfjörd JE
   19. Foekens JA
   20. Greaves M
   21. Hosoda F
   22. Hutter B
   23. Ilicic T
   24. Imbeaud S
   25. Imielinski M
   26. Imielinsk M
   27. Jäger N
   28. Jones DT
   29. Jones D
   30. Knappskog S
   31. Kool M
   32. Lakhani SR
   33. López-Otín C
   34. Martin S
   35. Munshi NC
   36. Nakamura H
   37. Northcott PA
   38. Pajic M
   39. Papaemmanuil E
   40. Paradiso A
   41. Pearson JV
   42. Puente XS
   43. Raine K
   44. Ramakrishna M
   45. Richardson AL
   46. Richter J
   47. Rosenstiel P
   48. Schlesner M
   49. Schumacher TN
   50. Span PN
   51. Teague JW
   52. Totoki Y
   53. Tutt AN
   54. Valdés-Mas R
   55. van Buuren MM
   56. van 't Veer L
   57. Vincent-Salomon A
   58. Waddell N
   59. Yates LR
   60. Zucman-Rossi J
   61. Futreal PA
   62. McDermott U
   63. Lichter P
   64. Meyerson M
   65. Grimmond SM
   66. Siebert R
   67. Campo E
   68. Shibata T
   69. Pfister SM
   70. Campbell PJ
   71. Stratton MR
   72. Australian Pancreatic Cancer Genome Initiative
   73. ICGC Breast Cancer Consortium
   74. ICGC MMML-Seq Consortium
   75. ICGC PedBrain

   (2013) Signatures of mutational processes in human cancer

   Nature 500:415–421.

   https://doi.org/10.1038/nature12477

   • Google Scholar
2. 1. Alexandrov LB
   2. Jones PH
   3. Wedge DC
   4. Sale JE
   5. Campbell PJ
   6. Nik-Zainal S
   7. Stratton MR
   8. Peter J

   (2015) Clock-like mutational processes in human somatic cells

   Nature Genetics 47:1402–1407.

   https://doi.org/10.1038/ng.3441

   • Google Scholar
3. 1. Ames BN
   2. Shigenaga MK
   3. Hagen TM

   (1993) Oxidants, antioxidants, and the degenerative diseases of aging

   Proceedings of the National Academy of Sciences of the United States of America 90:7915–7922.

   https://doi.org/10.1073/pnas.90.17.7915

   • Google Scholar
4. 1. Bachman M
   2. Uribe-Lewis S
   3. Yang X
   4. Williams M
   5. Murrell A
   6. Balasubramanian S

   (2014) 5-Hydroxymethylcytosine is a predominantly stable DNA modification

   Nature Chemistry 6:1049–1055.

   https://doi.org/10.1038/nchem.2064

   • Google Scholar
5. 1. Booth MJ
   2. Marsico G
   3. Bachman M
   4. Beraldi D
   5. Balasubramanian S

   (2014) Quantitative sequencing of 5-formylcytosine in DNA at single-base resolution

   Nature Chemistry 6:435–440.

   https://doi.org/10.1038/nchem.1893

   • Google Scholar
6. 1. Brazauskas P
   2. Kriaucionis S

   (2014) DNA modifications: Another stable base in DNA

   Nature Chemistry 6:1031–1033.

   https://doi.org/10.1038/nchem.2115

   • Google Scholar
7. 1. Chen K
   2. Zhang J
   3. Guo Z
   4. Ma Q
   5. Xu Z
   6. Zhou Y
   7. Xu Z
   8. Li Z
   9. Liu Y
   10. Ye X
   11. Li X
   12. Yuan B
   13. Ke Y
   14. He C
   15. Zhou L
   16. Liu J
   17. Ci W

   (2016) Loss of 5-hydroxymethylcytosine is linked to gene body hypermethylation in kidney cancer

   Cell Research 26:103–118.

   https://doi.org/10.1038/cr.2015.150

   • Google Scholar
8. 1. Cortellino S
   2. Xu J
   3. Sannai M
   4. Moore R
   5. Caretti E
   6. Cigliano A
   7. Le Coz M
   8. Devarajan K
   9. Wessels A
   10. Soprano D
   11. Abramowitz LK
   12. Bartolomei MS
   13. Rambow F
   14. Bassi MR
   15. Bruno T
   16. Fanciulli M
   17. Renner C
   18. Klein-Szanto AJ
   19. Matsumoto Y
   20. Kobi D
   21. Davidson I
   22. Alberti C
   23. Larue L
   24. Bellacosa A

   (2011) Thymine DNA glycosylase is essential for active DNA demethylation by linked deamination-base excision repair

   Cell 146:67–79.

   https://doi.org/10.1016/j.cell.2011.06.020

   • Google Scholar
9. 1. Globisch D
   2. Münzel M
   3. Müller M
   4. Michalakis S
   5. Wagner M
   6. Koch S
   7. Brückl T
   8. Biel M
   9. Carell T

   (2010) Tissue distribution of 5-hydroxymethylcytosine and search for active demethylation intermediates

   PloS One 5:e15367.

   https://doi.org/10.1371/journal.pone.0015367

   • Google Scholar
10. 1. Guisan A
    2. Zimmermann NE

    (2000) Predictive habitat distribution models in ecology

    Ecological Modelling 135:147–186.

    https://doi.org/10.1016/S0304-3800(00)00354-9

    • Google Scholar
11. 1. Guo JU
    2. Su Y
    3. Zhong C
    4. Ming GL
    5. Song H

    (2011) Hydroxylation of 5-methylcytosine by TET1 promotes active DNA demethylation in the adult brain

    Cell 145:423–434.

    https://doi.org/10.1016/j.cell.2011.03.022

    • Google Scholar
12. 1. Hardeland U
    2. Bentele M
    3. Jiricny J
    4. Schär P

    (2003) The versatile thymine DNA-glycosylase: a comparative characterization of the human, Drosophila and fission yeast orthologs

    Nucleic Acids Research 31:2261–2271.

    https://doi.org/10.1093/nar/gkg344

    • Google Scholar
13. 1. Hashimoto H
    2. Zhang X
    3. Cheng X

    (2012) Excision of thymine and 5-hydroxymethyluracil by the MBD4 DNA glycosylase domain: structural basis and implications for active DNA demethylation

    Nucleic Acids Research 40:8276–8284.

    https://doi.org/10.1093/nar/gks628

    • Google Scholar
14. 1. He Y-F
    2. Li B-Z
    3. Li Z
    4. Liu P
    5. Wang Y
    6. Tang Q
    7. Ding J
    8. Jia Y
    9. Chen Z
    10. Li L
    11. Sun Y
    12. Li X
    13. Dai Q
    14. Song C-X
    15. Zhang K
    16. He C
    17. Xu G-L

    (2011) Tet-Mediated Formation of 5-Carboxylcytosine and Its Excision by TDG in Mammalian DNA

    Science 333:1303–1307.

    https://doi.org/10.1126/science.1210944

    • Google Scholar
15. 1. Hu X
    2. Zhang L
    3. Mao SQ
    4. Li Z
    5. Chen J
    6. Zhang RR
    7. Wu HP
    8. Gao J
    9. Guo F
    10. Liu W
    11. Xu GF
    12. Dai HQ
    13. Shi YG
    14. Li X
    15. Hu B
    16. Tang F
    17. Pei D
    18. Xu GL

    (2014) Tet and TDG mediate DNA demethylation essential for mesenchymal-to-epithelial transition in somatic cell reprogramming

    Cell Stem Cell 14:512–522.

    https://doi.org/10.1016/j.stem.2014.01.001

    • Google Scholar
16. 1. Inoue A
    2. Zhang Y

    (2011) Replication-Dependent Loss of 5-Hydroxymethylcytosine in Mouse Preimplantation Embryos

    Science 334:194.

    https://doi.org/10.1126/science.1212483

    • Google Scholar
17. 1. Jones PA

    (2012) Functions of DNA methylation: islands, start sites, gene bodies and beyond

    Nature Reviews. Genetics 13:484–492.

    https://doi.org/10.1038/nrg3230

    • Google Scholar
18. 1. Kemmerich K
    2. Dingler FA
    3. Rada C
    4. Neuberger MS

    (2012) Germline ablation of SMUG1 DNA glycosylase causes loss of 5-hydroxymethyluracil- and UNG-backup uracil-excision activities and increases cancer predisposition of Ung-/-Msh2-/- mice

    Nucleic Acids Research 40:6016–6025.

    https://doi.org/10.1093/nar/gks259

    • Google Scholar
19. 1. Klose RJ
    2. Bird AP

    (2006) Genomic DNA methylation: the mark and its mediators

    Trends in Biochemical Sciences 31:89–97.

    https://doi.org/10.1016/j.tibs.2005.12.008

    • Google Scholar
20. 1. Koren A
    2. Polak P
    3. Nemesh J
    4. Michaelson JJ
    5. Sebat J
    6. Sunyaev SR
    7. McCarroll SA
    8. a S

    (2012) Differential relationship of DNA replication timing to different forms of human mutation and variation

    American Journal of Human Genetics 91:1033–1040.

    https://doi.org/10.1016/j.ajhg.2012.10.018

    • Google Scholar
21. 1. Kriaucionis S
    2. Heintz N

    (2009) The nuclear DNA base 5-Hydroxymethylcytosine is present in purkinje neurons and the brain

    Science 324:929–930.

    https://doi.org/10.1126/science.1169786

    • Google Scholar
22. 1. Krueger F
    2. Kreck B
    3. Franke A
    4. Andrews SR

    (2012) DNA methylome analysis using short bisulfite sequencing data

    Nature Methods 9:145–151.

    https://doi.org/10.1038/nmeth.1828

    • Google Scholar
23. 1. Lawrence MS
    2. Stojanov P
    3. Polak P
    4. Kryukov GV
    5. Cibulskis K
    6. Sivachenko A
    7. Carter SL
    8. Stewart C
    9. Mermel CH
    10. Roberts SA
    11. Kiezun A
    12. Hammerman PS
    13. McKenna A
    14. Drier Y
    15. Zou L
    16. Ramos AH
    17. Pugh TJ
    18. Stransky N
    19. Helman E
    20. Kim J
    21. Sougnez C
    22. Ambrogio L
    23. Nickerson E
    24. Shefler E
    25. Cortés ML
    26. Auclair D
    27. Saksena G
    28. Voet D
    29. Noble M
    30. DiCara D
    31. Lin P
    32. Lichtenstein L
    33. Heiman DI
    34. Fennell T
    35. Imielinski M
    36. Hernandez B
    37. Hodis E
    38. Baca S
    39. Dulak AM
    40. Lohr J
    41. Landau DA
    42. Wu CJ
    43. Melendez-Zajgla J
    44. Hidalgo-Miranda A
    45. Koren A
    46. McCarroll SA
    47. Mora J
    48. Lee RS
    49. Crompton B
    50. Onofrio R
    51. Parkin M
    52. Winckler W
    53. Ardlie K
    54. Gabriel SB
    55. Roberts CW
    56. Biegel JA
    57. Stegmaier K
    58. Bass AJ
    59. Garraway LA
    60. Meyerson M
    61. Golub TR
    62. Gordenin DA
    63. Sunyaev S
    64. Lander ES
    65. Getz G

    (2013) Mutational heterogeneity in cancer and the search for new cancer-associated genes

    Nature 499:214–218.

    https://doi.org/10.1038/nature12213

    • Google Scholar
24. 1. Li E
    2. Bestor TH
    3. Jaenisch R

    (1992) Targeted mutation of the DNA methyltransferase gene results in embryonic lethality

    Cell 69:915–926.

    https://doi.org/10.1016/0092-8674(92)90611-F

    • Google Scholar
25. 1. Li W
    2. Liu M

    (2011) Distribution of 5-Hydroxymethylcytosine in Different human tissues

    Journal of Nucleic Acids 2011:1–5.

    https://doi.org/10.4061/2011/870726

    • Google Scholar
26. 1. Lindahl T
    2. Nyberg B

    (1974) Heat-induced deamination of cytosine residues in deoxyribonucleic acid

    Biochemistry 13:3405–3410.

    https://doi.org/10.1021/bi00713a035

    • Google Scholar
27. 1. Lister R
    2. Mukamel EA
    3. Nery JR
    4. Urich M
    5. Puddifoot CA
    6. Johnson ND
    7. Lucero J
    8. Huang Y
    9. Dwork AJ
    10. Schultz MD
    11. Yu M
    12. Tonti-Filippini J
    13. Heyn H
    14. Hu S
    15. Wu JC
    16. Rao A
    17. Esteller M
    18. He C
    19. Haghighi FG
    20. Sejnowski TJ
    21. Behrens MM
    22. Ecker JR

    (2013) Global epigenomic reconfiguration during mammalian brain development

    Science 341:1237905.

    https://doi.org/10.1126/science.1237905

    • Google Scholar
28. 1. Maiti A
    2. Drohat AC

    (2011) Thymine DNA glycosylase can rapidly excise 5-formylcytosine and 5-carboxylcytosine: potential implications for active demethylation of CpG sites

    The Journal of Biological Chemistry 286:35334–35338.

    https://doi.org/10.1074/jbc.C111.284620

    • Google Scholar
29. 1. Mellén M
    2. Ayata P
    3. Dewell S
    4. Kriaucionis S
    5. Heintz N

    (2012) MeCP2 binds to 5hmC enriched within active genes and accessible chromatin in the nervous system

    Cell 151:1417–1430.

    https://doi.org/10.1016/j.cell.2012.11.022

    • Google Scholar
30. 1. Mittlböck M
    2. Heinzl H

    (2004)

    Proceedings of The 1st European Workshop on the Assessment of Diagnostic Performance

    71–80, Pseudo R-squared measures for generalized linear models, Proceedings of The 1st European Workshop on the Assessment of Diagnostic Performance, Italy, Milan.

    • Google Scholar
31. 1. Moréra S
    2. Grin I
    3. Vigouroux A
    4. Couvé S
    5. Henriot V
    6. Saparbaev M
    7. Ishchenko AA

    (2012) Biochemical and structural characterization of the glycosylase domain of MBD4 bound to thymine and 5-hydroxymethyuracil-containing DNA

    Nucleic Acids Research 40:9917–9926.

    https://doi.org/10.1093/nar/gks714

    • Google Scholar
32. 1. Nabel CS
    2. Jia H
    3. Ye Y
    4. Shen L
    5. Goldschmidt HL
    6. Stivers JT
    7. Zhang Y
    8. Kohli RM

    (2012) AID/APOBEC deaminases disfavor modified cytosines implicated in DNA demethylation

    Nature Chemical Biology 8:751–758.

    https://doi.org/10.1038/nchembio.1042

    • Google Scholar
33. 1. Nestor CE
    2. Ottaviano R
    3. Reddington J
    4. Sproul D
    5. Reinhardt D
    6. Dunican D
    7. Katz E
    8. Dixon JM
    9. Harrison DJ
    10. Meehan RR

    (2012) Tissue type is a major modifier of the 5-hydroxymethylcytosine content of human genes

    Genome Research 22:467–477.

    https://doi.org/10.1101/gr.126417.111

    • Google Scholar
34. 1. Nik-Zainal S
    2. Alexandrov LB
    3. Wedge DC
    4. Van Loo P
    5. Greenman CD
    6. Raine K
    7. Jones D
    8. Hinton J
    9. Marshall J
    10. Stebbings LA
    11. Menzies A
    12. Martin S
    13. Leung K
    14. Chen L
    15. Leroy C
    16. Ramakrishna M
    17. Rance R
    18. Lau KW
    19. Mudie LJ
    20. Varela I
    21. McBride DJ
    22. Bignell GR
    23. Cooke SL
    24. Shlien A
    25. Gamble J
    26. Whitmore I
    27. Maddison M
    28. Tarpey PS
    29. Davies HR
    30. Papaemmanuil E
    31. Stephens PJ
    32. McLaren S
    33. Butler AP
    34. Teague JW
    35. Jönsson G
    36. Garber JE
    37. Silver D
    38. Miron P
    39. Fatima A
    40. Boyault S
    41. Langerød A
    42. Tutt A
    43. Martens JW
    44. Aparicio SA
    45. Borg Å
    46. Salomon AV
    47. Thomas G
    48. Børresen-Dale AL
    49. Richardson AL
    50. Neuberger MS
    51. Futreal PA
    52. Campbell PJ
    53. Stratton MR
    54. Breast Cancer Working Group of the International Cancer Genome Consortium

    (2012) Mutational processes molding the genomes of 21 breast cancers

    Cell 149:979–993.

    https://doi.org/10.1016/j.cell.2012.04.024

    • Google Scholar
35. 1. Nilsen H
    2. Haushalter KA
    3. Robins P
    4. Barnes DE
    5. Verdine GL
    6. Lindahl T

    (2001) Excision of deaminated cytosine from the vertebrate genome: role of the SMUG1 uracil-DNA glycosylase

    The EMBO Journal 20:4278–4286.

    https://doi.org/10.1093/emboj/20.15.4278

    • Google Scholar
36. 1. Okano M
    2. Bell DW
    3. Haber DA
    4. Li E

    (1999) DNA methyltransferases Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian development

    Cell 99:247–257.

    https://doi.org/10.1016/S0092-8674(00)81656-6

    • Google Scholar
37. 1. Pacis A
    2. Tailleux L
    3. Morin AM
    4. Lambourne J
    5. MacIsaac JL
    6. Yotova V
    7. Dumaine A
    8. Danckaert A
    9. Luca F
    10. Grenier JC
    11. Hansen KD
    12. Gicquel B
    13. Yu M
    14. Pai A
    15. He C
    16. Tung J
    17. Pastinen T
    18. Kobor MS
    19. Pique-Regi R
    20. Gilad Y
    21. Barreiro LB

    (2015) Bacterial infection remodels the DNA methylation landscape of human dendritic cells

    Genome Research 25:1801–1811.

    https://doi.org/10.1101/gr.192005.115

    • Google Scholar
38. 1. Pleasance ED
    2. Stephens PJ
    3. O'Meara S
    4. McBride DJ
    5. Meynert A
    6. Jones D
    7. Lin ML
    8. Beare D
    9. Lau KW
    10. Greenman C
    11. Varela I
    12. Nik-Zainal S
    13. Davies HR
    14. Ordoñez GR
    15. Mudie LJ
    16. Latimer C
    17. Edkins S
    18. Stebbings L
    19. Chen L
    20. Jia M
    21. Leroy C
    22. Marshall J
    23. Menzies A
    24. Butler A
    25. Teague JW
    26. Mangion J
    27. Sun YA
    28. McLaughlin SF
    29. Peckham HE
    30. Tsung EF
    31. Costa GL
    32. Lee CC
    33. Minna JD
    34. Gazdar A
    35. Birney E
    36. Rhodes MD
    37. McKernan KJ
    38. Stratton MR
    39. Futreal PA
    40. Campbell PJ

    (2010) A small-cell lung cancer genome with complex signatures of tobacco exposure

    Nature 463:184–190.

    https://doi.org/10.1038/nature08629

    • Google Scholar
39. 1. Poon SL
    2. Pang ST
    3. McPherson JR
    4. Yu W
    5. Huang KK
    6. Guan P
    7. Weng WH
    8. Siew EY
    9. Liu Y
    10. Heng HL
    11. Chong SC
    12. Gan A
    13. Tay ST
    14. Lim WK
    15. Cutcutache I
    16. Huang D
    17. Ler LD
    18. Nairismägi ML
    19. Lee MH
    20. Chang YH
    21. Yu KJ
    22. Chan-On W
    23. Li BK
    24. Yuan YF
    25. Qian CN
    26. Ng KF
    27. Wu CF
    28. Hsu CL
    29. Bunte RM
    30. Stratton MR
    31. Futreal PA
    32. Sung WK
    33. Chuang CK
    34. Ong CK
    35. Rozen SG
    36. Tan P
    37. Teh BT

    (2013) Genome-wide mutational signatures of aristolochic acid and its application as a screening tool

    Science Translational Medicine 5:197ra101.

    https://doi.org/10.1126/scitranslmed.3006086

    • Google Scholar
40. 1. Rangam G
    2. Schmitz KM
    3. Cobb AJ
    4. Petersen-Mahrt SK

    (2012) AID enzymatic activity is inversely proportional to the size of cytosine C5 orbital cloud

    PLoS ONE 7:e43279.

    https://doi.org/10.1371/journal.pone.0043279

    • Google Scholar
41. 1. Rasmussen KD
    2. Helin K

    (2016) Role of TET enzymes in DNA methylation, development, and cancer

    Genes & Development 30:733–750.

    https://doi.org/10.1101/gad.276568.115

    • Google Scholar
42. 1. Saunders CT
    2. Wong WS
    3. Swamy S
    4. Becq J
    5. Murray LJ
    6. Cheetham RK

    (2012) Strelka: accurate somatic small-variant calling from sequenced tumor-normal sample pairs

    Bioinformatics 28:1811–1817.

    https://doi.org/10.1093/bioinformatics/bts271

    • Google Scholar
43. 1. Schultz MD
    2. He Y
    3. Whitaker JW
    4. Hariharan M
    5. Mukamel EA
    6. Leung D
    7. Rajagopal N
    8. Nery JR
    9. Urich MA
    10. Chen H
    11. Lin S
    12. Lin Y
    13. Jung I
    14. Schmitt AD
    15. Selvaraj S
    16. Ren B
    17. Sejnowski TJ
    18. Wang W
    19. Ecker JR

    (2015) Human body epigenome maps reveal noncanonical DNA methylation variation

    Nature 523:212–216.

    https://doi.org/10.1038/nature14465

    • Google Scholar
44. 1. Schuster-Böckler B
    2. Lehner B

    (2012) Chromatin organization is a major influence on regional mutation rates in human cancer cells

    Nature 488:504–507.

    https://doi.org/10.1038/nature11273

    • Google Scholar
45. 1. Shen L
    2. Wu H
    3. Diep D
    4. Yamaguchi S
    5. D'Alessio AC
    6. Fung HL
    7. Zhang K
    8. Zhang Y

    (2013) Genome-wide analysis reveals TET- and TDG-dependent 5-methylcytosine oxidation dynamics

    Cell 153:692–706.

    https://doi.org/10.1016/j.cell.2013.04.002

    • Google Scholar
46. 1. Spruijt CG
    2. Gnerlich F
    3. Smits AH
    4. Pfaffeneder T
    5. Jansen PW
    6. Bauer C
    7. Münzel M
    8. Wagner M
    9. Müller M
    10. Khan F
    11. Eberl HC
    12. Mensinga A
    13. Brinkman AB
    14. Lephikov K
    15. Müller U
    16. Walter J
    17. Boelens R
    18. van Ingen H
    19. Leonhardt H
    20. Carell T
    21. Vermeulen M

    (2013) Dynamic readers for 5-(hydroxy)methylcytosine and its oxidized derivatives

    Cell 152:1146–1159.

    https://doi.org/10.1016/j.cell.2013.02.004

    • Google Scholar
47. 1. Stephens PJ
    2. Tarpey PS
    3. Davies H
    4. Van Loo P
    5. Greenman C
    6. Wedge DC
    7. Nik-Zainal S
    8. Martin S
    9. Varela I
    10. Bignell GR
    11. Yates LR
    12. Papaemmanuil E
    13. Beare D
    14. Butler A
    15. Cheverton A
    16. Gamble J
    17. Hinton J
    18. Jia M
    19. Jayakumar A
    20. Jones D
    21. Latimer C
    22. Lau KW
    23. McLaren S
    24. McBride DJ
    25. Menzies A
    26. Mudie L
    27. Raine K
    28. Rad R
    29. Chapman MS
    30. Teague J
    31. Easton D
    32. Langerød A
    33. Lee MT
    34. Shen CY
    35. Tee BT
    36. Huimin BW
    37. Broeks A
    38. Vargas AC
    39. Turashvili G
    40. Martens J
    41. Fatima A
    42. Miron P
    43. Chin SF
    44. Thomas G
    45. Boyault S
    46. Mariani O
    47. Lakhani SR
    48. van de Vijver M
    49. van 't Veer L
    50. Foekens J
    51. Desmedt C
    52. Sotiriou C
    53. Tutt A
    54. Caldas C
    55. Reis-Filho JS
    56. Aparicio SA
    57. Salomon AV
    58. Børresen-Dale AL
    59. Richardson AL
    60. Campbell PJ
    61. Futreal PA
    62. Stratton MR
    63. Zainal SN
    64. Oslo Breast Cancer Consortium (OSBREAC)

    (2012) The landscape of cancer genes and mutational processes in breast cancer

    Nature 486:400–404.

    https://doi.org/10.1038/nature11017

    • Google Scholar
48. 1. Supek F
    2. Lehner B
    3. Hajkova P
    4. Warnecke T

    (2014) Hydroxymethylated cytosines are associated with elevated C to G transversion rates

    PLoS Genetics 10:e1004585.

    https://doi.org/10.1371/journal.pgen.1004585

    • Google Scholar
49. 1. Szwagierczak A
    2. Bultmann S
    3. Schmidt CS
    4. Spada F
    5. Leonhardt H

    (2010) Sensitive enzymatic quantification of 5-hydroxymethylcytosine in genomic DNA

    Nucleic Acids Research 38:e181.

    https://doi.org/10.1093/nar/gkq684

    • Google Scholar
50. 1. Tahiliani M
    2. Koh KP
    3. Shen Y
    4. Pastor WA
    5. Bandukwala H
    6. Brudno Y
    7. Agarwal S
    8. Iyer LM
    9. Liu DR
    10. Aravind L
    11. Rao A

    (2009) Conversion of 5-Methylcytosine to 5-Hydroxymethylcytosine in mammalian DNA by MLL partner TET1

    Science 324:930–935.

    https://doi.org/10.1126/science.1170116

    • Google Scholar
51. 1. Takai H
    2. Masuda K
    3. Sato T
    4. Sakaguchi Y
    5. Suzuki T
    6. Suzuki T
    7. Koyama-Nasu R
    8. Nasu-Nishimura Y
    9. Katou Y
    10. Ogawa H
    11. Morishita Y
    12. Kozuka-Hata H
    13. Oyama M
    14. Todo T
    15. Ino Y
    16. Mukasa A
    17. Saito N
    18. Toyoshima C
    19. Shirahige K
    20. Akiyama T

    (2014) 5-Hydroxymethylcytosine plays a critical role in glioblastomagenesis by recruiting the CHTOP-methylosome complex

    Cell Reports 9:48–60.

    https://doi.org/10.1016/j.celrep.2014.08.071

    • Google Scholar
52. 1. Taylor BJ
    2. Nik-Zainal S
    3. Wu YL
    4. Stebbings LA
    5. Raine K
    6. Campbell PJ
    7. Rada C
    8. Stratton MR
    9. Neuberger MS

    (2013) DNA deaminases induce break-associated mutation showers with implication of APOBEC3B and 3A in breast cancer kataegis

    eLife 2:e00534.

    https://doi.org/10.7554/eLife.00534

    • Google Scholar
53. 1. Tomasetti C
    2. Vogelstein B
    3. Parmigiani G

    (2013) Half or more of the somatic mutations in cancers of self-renewing tissues originate prior to tumor initiation

    Proceedings of the National Academy of Sciences of the United States of America 110:1999–2004.

    https://doi.org/10.1073/pnas.1221068110

    • Google Scholar
54. 1. Tomkova M
    2. McClellan M
    3. Kriaucionis S
    4. Schuster-Boeckler B

    (2016) 5-hydroxymethylcytosine marks regions with reduced mutation frequency in human DNA

    eLife 5:e17082.

    https://doi.org/10.7554/eLife.17082

    • Google Scholar
55. 1. Visvader JE

    (2011) Cells of origin in cancer

    Nature 469:314–322.

    https://doi.org/10.1038/nature09781

    • Google Scholar
56. 1. Wang K
    2. Yuen ST
    3. Xu J
    4. Lee SP
    5. Yan HH
    6. Shi ST
    7. Siu HC
    8. Deng S
    9. Chu KM
    10. Law S
    11. Chan KH
    12. Chan AS
    13. Tsui WY
    14. Ho SL
    15. Chan AK
    16. Man JL
    17. Foglizzo V
    18. Ng MK
    19. Chan AS
    20. Ching YP
    21. Cheng GH
    22. Xie T
    23. Fernandez J
    24. Li VS
    25. Clevers H
    26. Rejto PA
    27. Mao M
    28. Leung SY

    (2014) Whole-genome sequencing and comprehensive molecular profiling identify new driver mutations in gastric cancer

    Nature Genetics 46:573–582.

    https://doi.org/10.1038/ng.2983

    • Google Scholar
57. 1. Wen L
    2. Li X
    3. Yan L
    4. Tan Y
    5. Li R
    6. Zhao Y
    7. Wang Y
    8. Xie J
    9. Zhang Y
    10. Song C
    11. Yu M
    12. Liu X
    13. Li X
    14. Hou Y
    15. Guo H
    16. Wu X
    17. He C
    18. He C
    19. Tang F
    20. Tang F
    21. Qiao J

    (2014) Whole-genome analysis of 5-hydroxymethylcytosine and 5-methylcytosine at base resolution in the human brain

    Genome Biology 15:R49.

    https://doi.org/10.1186/gb-2014-15-3-r49

    • Google Scholar
58. 1. Wossidlo M
    2. Nakamura T
    3. Lepikhov K
    4. Marques CJ
    5. Zakhartchenko V
    6. Boiani M
    7. Arand J
    8. Nakano T
    9. Reik W
    10. Walter J

    (2011) 5-Hydroxymethylcytosine in the mammalian zygote is linked with epigenetic reprogramming

    Nature Communications 2:241.

    https://doi.org/10.1038/ncomms1240

    • Google Scholar
59. 1. Wu H
    2. Zhang Y

    (2011) Tet1 and 5-hydroxymethylation: a genome-wide view in mouse embryonic stem cells

    Cell Cycle 10:2428–2436.

    https://doi.org/10.4161/cc.10.15.16930

    • Google Scholar
60. 1. Wu S
    2. Powers S
    3. Zhu W
    4. Hannun YA

    (2016) Substantial contribution of extrinsic risk factors to cancer development

    Nature 529:43–47.

    https://doi.org/10.1038/nature16166

    • Google Scholar
61. 1. Yu M
    2. Hon GC
    3. Szulwach KE
    4. Song CX
    5. Zhang L
    6. Kim A
    7. Li X
    8. Dai Q
    9. Shen Y
    10. Park B
    11. Min JH
    12. Jin P
    13. Ren B
    14. He C

    (2012) Base-resolution analysis of 5-hydroxymethylcytosine in the mammalian genome

    Cell 149:1368–1380.

    https://doi.org/10.1016/j.cell.2012.04.027

    • Google Scholar

Author details

1. Marketa Tomkova

   Ludwig Cancer Research Oxford, University of Oxford, Oxford, United Kingdom

   Contribution

   MT, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

2. Michael McClellan

   Ludwig Cancer Research Oxford, University of Oxford, Oxford, United Kingdom

   Contribution

   MM, Acquisition, analysis and interpretation of HPLC measurements

   Competing interests

   The authors declare that no competing interests exist.

3. Skirmantas Kriaucionis

   Ludwig Cancer Research Oxford, University of Oxford, Oxford, United Kingdom

   Contribution

   SK, Conception and design, Drafting or revising the article

   For correspondence

   skirmantas.kriaucionis@ludwig.ox.ac.uk

   Competing interests

   The authors declare that no competing interests exist.

4. Benjamin Schuster-Boeckler

   Ludwig Cancer Research Oxford, University of Oxford, Oxford, United Kingdom

   Contribution

   BS-B, Conception and design, Drafting or revising the article

   For correspondence

   benjamin.schuster-boeckler@ludwig.ox.ac.uk

   Competing interests

   The authors declare that no competing interests exist.

   "This ORCID iD identifies the author of this article:" 0000-0002-8892-5133

• 3,058

  views

• 717

  downloads

• 27

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.