Complex correlations: Replication Timing and Mutational Landscapes during Cancer and Genome Evolution

Abstract

A recent flurry of reports correlates replication timing(RT) with rates of mutation during both evolution and cancer. Specifically, point mutations and copy number losses correlate with late replication, while copy number gains and other rearrangements correlate with early replication. In some cases, plausible mechanisms have been proposed. Point mutation rates may reflect temporal variation in repair mechanisms. Transcription-induced double-strand breaks are expected to occur in transcriptionally active early replicating chromatin. Fusion partners are generally in close proximity, and chromatin in close proximity replicates at similar times. However, temporal enrichment of copy number gains and losses remains an enigma. Moreover, many conclusions are compromised by a lack of matched RT and sequence datasets, filtering out developmental variation in RT, and the use of cell lines to make inferences about germline evolution.

Introduction

Recent evidence has unveiled strong correlations among replication timing (RT), and various forms of genetic mutation in both the germline and cancer [1]. The RT program in mammalian cells is regulated at the level of large (400–800 kb) chromosomal segments (“replication domains”) that replicate at specific times during S phase, with approximately half of these segments replicating at significantly different times in different cell types [2]. For reasons that are still poorly understood there is a near-precise correlation between RT and megabase-scale chromatin interaction spatial compartments defined by genome-wide chromosome-conformation capture (Hi-C), implying that domains in close spatial proximity replicate at similar times [3–5]. Therefore, control of replication invokes mechanisms that are regulated in both space and time. Moreover, early vs late replicating compartments are correlated with active vs inactive transcription and open vs closed chromatin [6–8], which could contribute to varying rates of mutation in the different compartments. In this review, we will summarize recent findings on the correlations of RT to multiple types of mutagenic events in germline and cancer genomes. Potential mechanisms that directly or indirectly couple genome mutation with RT will be discussed.

Replication Timing and Base Substitutions

An initial study spanning 1% of the human genome aligned RT data to either human-primate substitutions or human SNP density, revealing increases of 22% and 53%, respectively, for progressively later replicating DNA [9]. This analysis was followed by genome-wide studies reporting similar results [10,11]. These studies used RT data from highly transformed HeLa cells but, more recently, this concern was addressed by comparing RT profiles from six immortalized lymphoblastoid cell lines derived from father-mother-offspring trios to their matched whole genome sequences. This study found a strikingly more abundant SNP density in late replicating regions, with increases of >2-fold and >6-fold in transition and transversion mutations, respectively [12]. Thus, the removal of population noise in this study may have revealed a more intimate relationship between RT and base substitutions. This correlation is not specific to primates. When polymorphism between 6 Drosophila species was compared to the RT of Kc cells (an embryonic cell line from D. melanogaster), up to 30% increase in divergence was found from early to late replicated regions [13].

There are two important caveats with using somatic cells to draw inferences about evolution. First, it is not possible to distinguish whether mutations arose in the germline (parent to offspring) or during somatic differentiation, since somatic tissue undergoes considerably more cell divisions than germline tissue, which is sequestered early in development. Second, most studies have used replication profiles from somatic cell lines, but the temporal order of replication is significantly different in germline vs. somatic cells [2,6,8,14]. In fact, alignment between divergence of intronic sequences from Mus musculus and Rattus norvegicus reference genomes with RT profiles from mouse embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs) revealed only a 10.5% increase from early to late S phase [15], suggesting that mutation rates measured in somatic cells or cell lines may not accurately reflect the germline.

The problem with comparing cell lines to draw conclusions about germline evolution is underscored by studies of various cancer cell types [16–18]. Accumulation of single-nucleotide variation (SNV) in late-replicating/nuclear lamina associated domains (LADs) has been reported during the development of drug resistance in lung cancer [16]. Comparison of SNV in the human germline vs. multiple cancer cell lines revealed a more drastic increase with RT for cancer cells [17,19]. Cancer cells show the same mutational distribution as the germline [17], suggesting that the increased mutational skew may nonetheless arise from similar mechanisms. An important caveat of these cancer cell studies is that, to avoid complications of developmental RT control, they discarded the RT-variant half of the genome from their analyses. Hence, aspects of cancer genome evolution shaped by abnormal developmental control or RT were ignored. Cancer subtypes and even individual cancer clones have their own RT patterns [20], so a complete understanding of the role of replication timing in cancer genome evolution will require studies that compare RT to genome sequences in the same cancer clones.

Altogether, while absolute mutation rates and RT vary by cell type and species, an overall increasing gradient of point mutation rates with later RT has been established in a wide variety of eukaryotic species and probably is conserved among all eukaryotic organisms. This trend is still significant after controlling for covariant features such as recombination hot spots [9], differing tri-nucleotide compositions, G-C content, distance to telomeres, meiotic crossover rate, chromatin compaction [10], and local recombination rates, suggesting (but by no means confirming) a more direct relationship to replication time [13]. Moreover, the somatic vs. germline argument does not apply to single-cell organisms. In Saccharomyces cerevisiae, synonymous SNP and intergenic SNP rates in different strains were found to be elevated >30% in late replicating regions [21]. In fact, late replication was shown to be both necessary and sufficient to cause a high level of point mutations of experimentally manipulated chromosome segments in S. cerevisiae [22].

Proposed Mechanisms for Increasing Point Mutation Rates in Late S Phase

In S. cerevisiae, late S phase templates tend to be repaired with error-prone repair mechanisms [23]. Rev1, in a complex with error-prone Polζ (Rev3/7), increases ~50 fold in late S to G2 phases and deletion of Rev1 significantly lowers mutation rate specifically in late-replicating regions. In contrast, replication forks stalled at lesions during early S phase are accurately repaired by either copying from the correct template or high-fidelity translesion synthesis (Figure 1) [23,24]. Since these proteins are highly conserved [25,26], this mechanism might be common to all eukaryotes. In multi-cellular eukaryotes, early replicating regions are usually gene-rich and transcriptionally active, and lesions in active transcription units tend to be repaired by high fidelity transcription-coupled nucleotide excision repair (Figure 1) [27]. Therefore, different repair mechanisms operating on early vs. late templates could also contribute to the observed trend. Moreover, the substitution rate of replication Timing Transition Regions (TTRs) between replication domains that replicate at significantly different times is higher when those TTRs are interacting more closely with late replicating chromatin [17]. This observation supports the “body-guard” hypothesis (Figure 1) that heterochromatin at the periphery of the nucleus absorbs mutagens to protect interior, more genetically active, chromatin [28]. Other speculative mechanisms have also been proposed. For example, earlier replicating sequences would have more time to be repaired by hypothetical mechanisms that could act after mismatch repair [9]. Also, lower or altered nucleotide pools in late S could trigger mis-incorporation [9], although there is little direct evidence [29].

Figure 1. Complex relationship between replication timing and genome variation.

In the G1-nucleus, heterochromatin, localized to the periphery of the nucleus, may form a first defense line for invading mutagens, increasing point mutation rates in late replicating chromatin. Higher levels of transcription in the interior euchromatin could facilitate transcription-coupled nucleotide excision repair (lowering point mutation rates), but could also promote DSBs leading to translocations. In early S phase, collisions between replication and transcription machineries could increase rates of translocations or duplications, which develop either within a domain (observed during reprogramming) or between domains (observed in cancers). In late-S phase, error-prone repair mechanisms cause accumulation of point mutations.

Replication Timing and Copy Number Variation (CNV)

A comparison of CNVs in D. melanogaster and D. simulans with RT from 2 D. melanogaster cell lines found that duplications are enriched in late regions, deletions in early regions [30,31]. No direct measurements of CNV across mammalian species have been reported, but analysis of 331,724 CNVs (in this study called somatic copy-number alterations, SCNAs) in 26 cancer types found the reverse trend to these fly studies [32]. This study demonstrated that the breakpoints for these large (~4.5 Mb) CNV gains are enriched in early replicating regions, whereas those for CNV losses are enriched in late replicating regions, although overall, most CNVs are in constitutively late replicating regions (Figure 1) [32]. Moreover, the boundaries for these large CNVs are of similar RT, and are in close spatial proximity in the nucleus [32]. RT and chromatin proximity together can predict the locations of approximately half of CNVs in the cancer genome [32]. Although these findings was challenged for a lack of statistical significance [33], an analysis using this same dataset successfully modeled CNV distribution using chromosome interaction maps [34]. These cancer studies, similar to the studies of point mutation discussed above, also discarded the RT-variant half of the genome and suffer from the same drawbacks.

A recent study directly examined the relationship between RT-variant domains and CNVs characteristics during human iPSC reprogramming (J. Lu and P. Lerou, submitted). They also found that CNV gains and losses occur preferentially in early and late replicating regions, respectively. Within the RT-variant regions, CNV gains were enriched in domains that replicate earlier in iPSCs as compared to fibroblasts, while losses were mostly in the constitutively late regions, a few of which replicate later in iPSCs than fibroblasts. This study underscores concerns with interpreting studies that remove the RT-variant portion of the genome, since this portion of the genome may have a unique signature of genetic abnormalities.

Proposed Mechanisms for CNV Gains and Losses Linked to Early and Late S Phase

How gains and losses are related to RT is still a matter of speculation. Replicative CNVs could arise from stalled or collapsed forks, which are prone to double-stranded DNA breaks (DSBs) [35,36]. DNA replication is regulated at the level of 400–800kb domains, within which several replicons are simultaneously activated. Since the distribution of origins differs in early and late replicating domains [37,38], the density of simultaneously active forks that could serve as recombination partners may vary. Moreover, regions replicating earlier have twice (or more in cases of re-replication) as many copies as late replicating regions for a significant fraction of the cell cycle, increasing the chance of recombination within vs. between domains [39]. These features of normal replication would tend to favor intra-domain CNVs (~10–300kb), which are more common during evolution and reprogramming ([12,30], J. Lu and P. Lerou, submitted). It seems likely that large inter-domain CNVs (500kb to Mega-base level), which are more commonly seen in cancers [40] arise by different mechanisms than smaller CNVs. Inter-domain CNVs are likely related to the spatial juxtaposition of similar RT domains. One proposed mechanism for large inter-domain CNVs is that domains with similar RT might share replication factories [34], but this is highly unlikely since independent domains have never been observed to share replication factories [41].

Replication-independent CNV mechanisms have also been invoked, including differences in homologous or non-homologous recombination activities [12] and DSB repair mechanisms in different chromatin contexts [42,43]. For example, a recent study has reported that HP1 antagonizes the effect of a histone lysine demethylase, which induces re-replication and thus amplification of specific sites [44], indicating that the HP1-enriched late compartment is suppressed from CNV gains by this mechanism.

Replication Timing, Chromosome Break Points and Translocations

Evolutionary breakpoints between mouse and human are enriched in early replicating regions, and fusion partners generally replicate at similar times in similar cell types from both species [5], consistent with the importance of spatial organization as invoked for CNV. Although these results are subject to the same concerns of using somatic cell RT data to infer evolutionary change, these tendencies are consistent with reports in cancer cells [45,46]. For example, in neuroblastoma cells, >50% of breakpoints mapped to early replicating regions, which is 3.7 times higher than late regions [46]. Additionally, the same trend was found in hematological cancer cells using high resolution RT datasets from a lymphoblastoid cell line [47]. Importantly, in these last two studies, RT data were derived from the same cell lines as translocation data, allowing more direct comparisons. Finally, in a study of acute lymphocytic leukemia, it was found that RT changes associated with translocations can be found in patient samples that lack the translocation [20], suggesting that RT may switch during carcinogenesis upstream of the translocation, placing translocation partners in spatial proximity.

Proposed Mechanisms for Higher Chromosome Break Frequency in Early S Phase

Longstanding cytogenetic evidence [48–56], recently confirmed with chromatin conformation capture [47,57–63], indicates that fusion partners are usually in close spatial proximity before the breaks occur. Since chromatin is organized in the nucleus such that domains of similar RT are in close proximity [4], there is a strong preference for partners to replicate at the same time [47], which is also reported in evolutionary fusion partners between mouse and human [5]. More recently, this concept was tested by experimentally inducing DSBs, revealing that both more common intra-chromosomal fusions and rare translocations were coupled to spatial proximity (Figure 1) [59,62].

The proximity concept explains why fusion partners have similar RT, but the preference for early replication may be due to the promotion of breaks by transcription. Translocations preferentially occur at genic regions, particularly at active transcription start sites [57,59] and translocation signatures in different cancers correlate with tissue-specific transcription profiles [47]. Transcription is a mutagenic process in that it creates topological stress, which leads to DSBs aggravated by topoisomerase inhibition [56]. Moreover, failure of RNA processing during transcription termination leads to genome instability via co-transcriptional R-loop mediated DSB formation [64]. Distal genes regulated by common transcription factors often share transcription factories (Figure 1), placing them in very close proximity under conditions conducive for non-homologous end joining [56,65,66]. In fact, transcriptional induction of TMPRSS2 and ERG (frequent translocation pairs in prostate cancer) results in their movement to a shared transcription factory and dramatically increases their rate of fusion upon exposure to irradiation [67]. Transcription-linked dsDNA breaks may be further exacerbated during replication by head-on collision of transcription and replication machineries (Figure 1) [68]. Since 75% of expressed genes are early replicating [7], transcription-mediated fusions can account for why translocations are enriched in early replicating regions.

Biological Significance of these Complex Correlations

Clearly, one can only speculate as to the significance of mutagenic biases observed in each replication compartment. The “body guard” hypothesis posits that, during somatic development, genes important for general cellular function need to be protected from mutation, while silent genes for alternate developmental lineages are dispensable [28]. From an evolutionary standpoint, late replicating genes could serve as a potential source for gene innovation, with increased point mutations facilitating more rapid evolution of cell-type specific late replicating genes [69]. This may also explain why most housekeeping genes essential for general cell functions exist in the early compartment [69].

Conclusions and Future Directions

It is clear that regions replicated at different times during S phase display distinct patterns of mutation (Figure 1). In yeast, late replication causes enhanced base substitution, possibly due to increased error-prone DNA repair activities in late S phase. A similar correlation is found in higher eukaryotes, but a clear causal relationship has not been established. In the case of CNV and translocations, both result from DSBs but likely involve different mechanisms since they show inverse correlation to RT. There is strong evidence for an involvement of transcription in translocations, which mainly occurs in early replicating regions. The mechanisms linking CNV to RT are more speculative. Both CNV and translocations occur between breakpoints of similar RT, suggesting the importance of sub-nuclear proximity for both events. Correlations between RT and the mutagenic landscape in the germline are consistent with findings in cancer, suggesting common underlying mechanisms. Many existing studies suffer from comparisons between genetic variation and RT in unmatched cell types. Since significant RT changes during embryogenesis [2,6,8,14] and carcinogenesis [20,70] have been well documented, and many high-resolution RT profiles now exist (www.replicationdomain.org and ENCODE), it is critical to analyze the mutations with RT datasets from matching cell types. This would also significantly improve the accuracy in predicting real cancer driver genes [71]. In sum, RT, repair, transcription, and the spatial organization of chromatin confine and refine the genome mutagenic landscapes. Further understanding of the exact molecular mechanism of the RT program and its relationship to genome architecture and transcription should shed light on mechanisms underlying the observed correlation between RT and genome mutational landscape.

Table 1.

Summary of literature

Type	Study	Species	Year	Source of RT Data	Sequence Data	Hi-C/Lamin alignment
Point mutation[9]	Evolution	Human-Primates	2009	1% of genome, HeLa[72]	Human/chimpanzee reference genomes	None
Point mutation[10]	Evolution	Human-Primates	2010	HeLa[10]	Human/primate/mouse reference genomes	None
Point mutation[15]	Evolution	Mouse	2010	mESC and miPSC[73]	Rat/mouse reference genome introns	None
Point mutation/CNV[12]	Evolution	Human	2012	6 lymphoblastoid[12]	WGS from father-mother-offspring trios	None
Point mutation[19]	Cancer/Evolution	Human	2012	4 cell types; constitutive E&L[2]	SNPs across 6 cancer types, 2 healthy genomes	None
Point mutation[17]	Cancer/Evolution	Human	2013	4 cell types; constitutive E&L[2]	Neutral sites across 5 cancer types, 2 healthy genomes	Hi-C
Point mutation[16]	Cancer	Human	2013	4 cell types; constitutive E&L[2]	Whole genome/exome of lung cancer cells	Lamin(human fibroblast)[76]
Point mutation[13]	Evolution	Fly	2011	Kc cells[14]	Synonymous sites in exon cores, putatively unconstrained intronic sites in 6 Drosophila species	None
Point mutation[21]	Evolution	Yeast	2011	S. cerevisiae[74]	Synonymous intergenic SNPs	None
CNV[32]	Cancer	Human	2011	4 cell types; constitutive E&L constitutive E&L[2]	CNVs across 26 cancer types cancer types (Average 4.5Mb )	Hi-C
CNV[31]	Evolution	Fly	2010	Kc and C18[14]	Duplications and deletions in 15 natural D. melanogaster populations	None
CNV[30]	Evolution	Fly	2011	Kc, Bg3 and S2 cells[14,75]	Duplications in 14 natural D. simulans lines	None
Break&Fusions[5]	Evolution	Human-mouse	2010	Fibroblasts and lymphoblasts[5]	Human and mouse reference genomes	Hi-C
Breaks[46]	Cancer	Human	2005	neuroblastoma and lymphoblastoid[46]	Breakpoints in neuroblastoma cell lines	None
Break&Fusions[47]	Cancer	Human	2012	GM06990[73]	Reciprocal chromosomal translocations in leukemia patients (TICdb )	Hi-C