Structural Variations in Cancer and the 3D Genome

Abstract

Structural variations (SVs) affect more of the cancer genome than any other type of somatic genetic alteration but difficulties in detecting and interpreting them have limited our understanding. Clinical cancer sequencing also increasingly aims to detect SVs, leading to a widespread necessity to interpret them. Recently, analyses of large whole-genome sequencing datasets revealed features that impact rates of SVs across the genome in different cancers. A striking feature has been the extent to which, in both their generation and their impacts on the selective fitness of cancer cells, SVs are more specific to individual cancer types than other genetic alterations such as single nucleotide variants. This review discusses how the folding of the 3D genome, and differences in its folding across cell types, affect observed SV rates in different cancers.

1. Importance and tissue-type specificity of cancer SVs

Among the three types of genetic alteration in the cancer genome – single nucleotide variants (SNVs), structural variants (SVs), and exogenous DNA – SVs affect by far the largest fraction of the cancer genome. SVs broadly represent two classes of alteration: whole-chromosome or genome-scale aneuploidies, and rearrangements between distant genomic loci. All copy-number alterations result from SVs, and the first three genetically targeted therapeutics in cancer targeted oncogenes activated by SVs: PML-RARA in the case of ATRA¹ amplified ERBB2 in the case of trastuzumab², and BCR-ABL1³ in the case of imatinib. All three of these SVs have a direct effect (gene fusion or amplification) on the DNA sequence of the oncogene, which spurred their detection and interpretation. While this approach was consequently expanded to develop targeted therapies against oncogenes with SNVs in their coding sequence, the vast majority of oncogenic SVs lie outside of the exome and remained mostly overlooked until now. With recent improvements in sequencing technologies (Box 1) and in our understanding of the epigenetic features that shape the three-dimensional genome, however, we are now detecting more SVs^4-6 and more oncogenic SVs^7-9. This is important both for cancer genome discovery--developing a better understanding of the oncogenic events that lead to cancer--and for clinical diagnostics, SVs can serve as important biomarkers to indicate patients who are likely to respond to targeted therapeutics. Indeed, whole-genome sequencing is now beginning to enter clinical practice ^10,11,12.

Box 1: Limitations to SV discovery and interpretation.

Whole-genome sequencing (WGS) has limited sensitivity for SV detection. Standard contemporary sequencing generates paired 100-250bp reads that represent the ends of longer (usually ~500 bp) fragments. SVs are indicated by “discordant reads”, whose paired ends align to genomic loci that are more distant than the typical fragment size, and by “split reads” that overlap SV breakpoints (Figure Box 1 panel A). Only part of a split read will align to each parental sequence, and these short segments are often unmappable. Complex SVs, with multiple breakpoints in close proximity, can further subdivide overlapping reads, increasing ambiguity.

Additional challenges involve untangling the linear genomic sequence generated by complex SVs (Figure Box 1 panel B). First, each individual SV in the complex event must be phased, determining which are on the same homologous chromosome. Second, the linear ordering of different SVs within a complex event can be ambiguous, especially when some SVs may have escaped detection. Complex SVs also often differ between cancer subclones, generating more ambiguity.

New computational and sequencing technologies are helping to solve these issues. Long-read¹⁵⁸, linked-read¹⁵⁹, and optical-mapping¹⁶⁰ generate reads or read sets that are more likely to contain unique sequences, reducing mapping ambiguity. Different SVs can be phased when they are captured on the same read pair or on separate reads that also contain sequences (such as germline variants) that are unique to a single haplotype (Figure Box 1 panel B). Single reads that capture two or more SVs can also topologically disambiguate the resulting linear sequence. SV detection methods based upon genome assembly^4,161,162 can also generate longer contigs that are more likely to align uniquely. Probabilistic methods are also being developed to integrate copy-number and SV data to reconstruct genome topologies^163-165, which sometimes disambiguates the topology of complex SVs¹⁶³. And orthogonal datasets can further aid SV detection. For example, Hi-C data can detect juxtapositions between normally distant genomic loci by indicating their three-dimensional proximity^166,167 and RNA sequencing can detect novel fusion transcripts^168-170.

Request for NRC graphics help:

1. Please improve the quality of the figure’s appearance. For example, color choices need to be better!

2. Please standardize appearance across subpanels and other figures

Figure Box 1.

Here, we focus on rearrangements rather than aneuploidies, and use “SV” to represent only the former. These SVs can be further classified as deletions, duplications, inversions, interchromosomal translocations, and insertions (Figure 1A), as well as any combination of these. As long as the SV remains isolated it can be fairly easy to assign it to one of these classes. However, a large share of SVs involve several simultaneous breaks and re-ligation. This creates complex SVs, which are harder to interpret (Figure 1B, Box 1).

Figure 1: Types of rearrangements.

A) Topological categories of simple SVs: translocations, deletions, duplications, inversions, and insertions. The initial DNA sequence is indicated in the center, with arrows pointing to sequences generated by each of these categories of SV. Underneath these SV-derived sequences are the copy-number profiles and SV calls that would be detected, projected onto the reference genome. In second-generation sequencing data, the copy-number profiles reflect the number of reads aligned at each locus. Here, “2” indicates the number of reads expected from loci with two copies. The SV calls are indicated by arrows connected by dotted lines. These represent the discordant and split reads in second-generation sequencing data, where part of the read maps to each side of the SV (the arrows). B) Mechanisms of generating complex SVs. These can originate in a single event (typical for chromothripsis) or sequence of events over multiple cell cycles (as in breakage-fusion-bridge cycles (BFBs) or with the accumulation and selection of double minutes). The resulting cancer genome as well as the pattern detected by second-generation sequencing are indicated analogous to (A). The detected patterns can be indicative of the SV generating process, (Amplification with inversions for BFB, focal high CN for DM, complex SVs across a chromosome with CN loss for chromothripsis) but are often overlayed by subsequent SVs in real cancer genomes. Additionally, one pattern of complex SVs can be mechanistically linked to another, increasing the probability of observing mixed patterns in real cancer genomes.

The figures for this paper were produced by Nature Reviews Cancer graphics artists and are under copyright to Nature Reviews Cancer. The figures can be viewed at : Dubois F, Sidiropoulos N, Weischenfeldt J, Beroukhim R. Structural variations in cancer and the 3D genome. Nature Reviews Cancer. 2022 September;22(9):533–546. PMID: 35764888; DOI: 10.1038/s41568-022-00488-9.

In contrast to SNVs, SVs can often affect multiple genes, often through resulting copy number aberrations, which affect one-third of the genome in the average cancer¹³. However, the vast majority of SVs occur outside of the small fraction of the genome that is represented in exome sequencing. Sensitive detection requires high coverage whole genome sequencing (WGS) – and even then, standard short-read sequencing can result in high false-negative rates, particularly for complex SVs and SVs in genomic loci that are difficult to map to the genome. However, recent advances in WGS, long- and linked-read sequencing, and availability of orthogonal epigenetic data are beginning to shed light on their prevalence and distribution (Box 1). Even after successful detection, the interpretation of SVs as driver vs passenger events has been complicated due to unknown rates at which SVs naturally occur across the genome due to mechanistic biases (their “background rate”). Additionally, even individual simple SVs can have a vast array of consequences including copy-number changes that extend over many megabases and simultaneous disruption of protein sequences and cis-regulatory elements, for example resulting in the simultaneous disruption of tumor suppressors and activation of oncogenes⁸. These same considerations, however--the manifold and extensive effects of individual SVs--add to the importance of understanding them. This is increasingly possible as we gain the ability to resolve and interpret the mechanistic and selective processes that determine where and why these alterations are observed. Only recently, however, have we begun to understand the role of epigenetics in shaping these mechanisms and the SVs that result^8,14,15.

While many SVs in germline DNA^16,17 have been linked to disease^18,19 the SV landscape in a typical cancer genome is distinctly influenced by somatic processes^5,20,21. Therefore, we will exclusively focus on somatically acquired SVs in this review. Germline SVs have recently been reviewed elsewhere²².

In contrast to SNVs, the rates at which SVs are observed in cancer appear to be more heavily influenced by the 3D genome⁵. The very nature of SVs, connecting distant loci, suggests such a relationship: the frequency with which genetic loci are fused by SVs is inversely proportional to the linear distance between them, which in the context of normal DNA has been modeled by a knot-free 3D conformation called a fractal globule^20,23,24. Their consequences are also often due in part to their effects on the 3D conformation of DNA. For example, driver fusion events, which rely largely on the hijacking of regulatory components to spur oncogene expression, have been observed in over a quarter of cancers²¹. In addition, extrachromosomal DNA (ecDNA), formed by SVs, has been detected in almost half of cancers^25,26 and involves changes in DNA conformation to bring tissue-type specific regulatory regions in close vicinity to oncogenes^15,27.

Perhaps for these reasons, specific 3D features of individual cell types appear to matter more for SVs. For example, whereas BRAF V600E mutations are frequently observed across many cancer types, oncogenic BRAF-KIAA1549 rearrangements have almost exclusively been observed in a single cancer type: juvenile pilocytic astrocytoma⁵. Likewise, whereas TMPRSS2-ERG rearrangements are common in prostate cancers, they are almost never observed in other cancers²⁸. The major genomic differences between the cell types that generate these cancers are epigenetic (meaning “molecules and mechanisms that can perpetuate alternative gene activity states in the context of the same DNA sequence”²⁹). Much of the epigenetic state of a cell is represented by its chromatin conformation and three-dimensional organization of DNA. This review describes mechanisms by which this 3D conformation of the genome affects rates at which SVs are observed–the way the 3D genome “shapes” SVs in cancer–including through its impact on rates of SV formation and on SV effects on cellular fitness.

2. An emerging understanding of the 3D genome

a. The pre-cancerous genome

Recently, methodologies have made it possible to infer the nuclear distance between any two loci in the genome. For example, sequencing-based methods such as Hi-C³⁰, a technique that assesses chromatin interactions by proximity-mediated ligation and sequencing, have now made it possible to study nuclear organization genome-wide at fine scale. It has also become possible to obtain microscopy-based distance-metrics for several regions simultaneously³¹, which can provide insights into the extent to which regions that are distant at the linear scale but close in nuclear distance tend to form SVs.

Analyses of these data have revealed general folding principles of the genome, which are important to understanding how its 3D structure impacts SV formation in cancer. Overall, the likelihood that two genomic loci are in contact is inversely proportional to the linear distance between them ²⁴. This has been attributed to a “fractal globule” (alternatively called “crumpled globule”) 3D conformation of DNA, which forms through local regions crumpling iteratively across each chromosome³². Hi-C has also revealed higher-resolution structures of DNA, including chromatin compartments and topologically associated domains (TADs), both of which affect the frequency of SVs in cancer.

The genome can be separated into large chromosome territories – areas occupied by particular chromosomes³³ – and smaller chromatin compartments into which they subdivide. When evaluating the latter, Rao et al identified ‘A’ and ‘B’ compartments corresponding to gene-rich, actively transcribed euchromatin associated with early replication timing, primarily located in the centre of the nucleus; and gene-poor, repressive, heterochromatin, preferentially located at the nuclear periphery (Figure 2A)^24,34-36. The separation of A and B compartments has been proposed to reflect phase separation between active and inactive chromatin based on their inherent physical properties³⁷,

Figure 2: Chromatin organization and its disruption by rearrangements in cancer.

A) The three-dimensional organization of DNA. Chromatin fibers can form regions of close interactions such as topologically associated domains (TADs). At the largest scale, chromatin is organized as a fractal globule, in which the likelihood that two sequences within a chromosome will be adjacent in three-dimensional space is proportional to the number of bases that separate them. B) TADs are highly interconnected DNA loops, as indicated by dense triangles in Hi-C maps, where higher interaction frequencies are represented by deeper red. Their boundaries are often defined by CTCF and cohesin binding, and genetic elements that regulate transcription (e.g. promoters, enhancers, and silencers) typically act upon genes within their TAD. C) Effects of SVs on TAD structure and characteristic resulting interaction maps. (top) Situation before the SV with two TADs separated by a boundary element. The ectopic contacts resulting from deletion (left), duplication (middle), and inversion (right) are shown in blue. As apparent in the interaction maps, all three can cause genes to interact with regulatory elements in the neighboring TAD. The reference genome projections (bottom) show characteristic interaction maps for each SV type, which enables their interpretation in Hi-C analysis. Figure panels modified from Spielmann M, Lupiáñez DG, Mundlos S. Nat Rev Genet. 2018.

The figures for this paper were produced by Nature Reviews Cancer graphics artists and are under copyright to Nature Reviews Cancer. The figures can be viewed at : Dubois F, Sidiropoulos N, Weischenfeldt J, Beroukhim R. Structural variations in cancer and the 3D genome. Nature Reviews Cancer. 2022 September;22(9):533–546. PMID: 35764888; DOI: 10.1038/s41568-022-00488-9.

TADs are chromatin domains that are typically 100 kb-1Mb in size and demarcated by architectural proteins including CCCTC binding factor (CTCF), and the cohesin complex (RAD21, SMC1, SMC3, and STAG1/2). TADs contain extensive chromatin looping such as promoter-enhancer interactions around these CTCF/cohesin “chromatin loop anchors”, that are likely to facilitate the regulation of gene expression^38-40. The orientation of the CTCF motif dictates the probability with which nearby loci are spatially juxtaposed. Hi-C data have shown that most CTCF-associated loops are in a convergent orientation (Figure 2B)^38,41-43, which is compatible with a loop-extrusion model, by which DNA is pulled through a ring-shaped structure composed of CTCF-Cohesin complex⁴⁴. Interestingly, recent studies have shown that a substantial proportion of enhancer-promoter loops occur independently of CTCF binding. The rules governing how such interactions occur are not well-understood

Although the 3D conformation of the genome is not equal to its epigenetic state, they are heavily interdependent. Histone marks have classically been used to identify epigenetic features such as enhancers, promoters, and open and closed chromatin⁴⁵. In Hi-C data, interaction loops appear as connections in 3D between elements such as enhancers and promoters⁴⁶. Chromatin regions represented as A and B compartments can also be identified in Hi-C data by their characteristic patterns²⁴ as well as using open chromatin data and methylation data ⁴⁷. For these reasons, our review describes both explicit relations between SVs and the 3D genome, and implicit relations indicated by relationships between SVs and epigenetic features of the genome.

b. Effects of SVs on the 3D genome

The 3D organization of the genome exhibits substantial and dynamic heterogeneity at the single-cell and even allelic level, which is shaped both by cellular states and stochasticity^48,49 and can be deregulated in cancer through SVs.

A major role of SVs in shaping the 3D genome is by altering interactions between CTCF-dependent and independent DNA loops. This can occur in three ways (Figure 2C). First, if a CTCF anchor is deleted, the two adjacent TADs will fuse. Second, duplications of CTCF anchors will create new TADs (“neo-TADs”). Third, inversions encompassing CTCF anchors will swap DNA territory between TADs. Translocations between chromosomes can either delete CTCF sites, thereby fusing TADs across chromosomes, or maintain CTCF sites, thereby swapping DNA territory between TADs on these different chromosomes. Dysregulation of CTCF leads to altered enhancer-promoter looping and gene expression in cancer^50,51.

SVs may also alter the global positioning of DNA territories within the nucleus. CRISPR engineering has shown that removal of cis-elements can lead to changes in compartment structure and DNA activation or repression^52,53 phenomena that have been associated with SVs in cancer⁵⁴ and reviewed in⁵⁵. DNA movement within the nucleus can also alter its relation to the nuclear membrane, thereby contributing to transcriptional repression reviewed in⁵⁶; it is interesting to speculate that adjacency to the nuclear membrane may also contribute to further DNA damage. Excess curvature of nuclear membranes in micronuclei and anaphase bridges has been linked to DNA damage and complex SVs^57-59

These effects of SVs in cancer genomes can modify transcription in multiple ways, including by altering copy-number and changing DNA compartment structure. The effects of SVs on chromatin looping also often modify interactions between enhancers, promoters, and other transcriptional regulatory elements, as discussed in detail below.

3. The two major determinants of observed rates of SVs in cancer: mechanistic biases and effects on evolutionary selection

The frequency with which SVs are observed in cancer is determined both by the rates at which they occur across cells in the body (mechanistic biases) and by their effects on cellular fitness, which lead cancer genetic studies to observe them more or less frequently than this underlying rate (selection biases). UnderstandingUnderstanding this better requires an appreciation of how genetic alterations arise and are enriched during cancer development, which we discuss below. In the following section, we will see how each of these factors, mechanism and cellular fitness, are influenced by the 3D conformation of the genome.

Cancer originates when a normal cell acquires a set of genetic alterations that lead it to grow uncontrollably⁶⁰. These oncogenic, or “driver”, alterations arise through a random process of mutagenesis that typically also generates genetic alterations that are either evolutionarily neutral or disadvantageous (Figure 3A). In a simplistic view, tumor cells will continue to grow if the net sum of their mutations on cellular fitness is positive. As a consequence, cells usually carry both neutral and even disadvantageous alterations, maintained as “passenger events”--and indeed these may outnumber the driver events by orders of magnitude due to the far greater numbers of genetic alterations with neutral or negative fitness effects. Both driver and passenger events include SNVs, SVs, and sometimes exogenous nucleic acids (e.g. introduced by oncogenic viruses). The driver events are oncogenic either due to changes to the protein-coding sequence of oncogenes or tumor suppressors or altered expression of the encoded proteins (or both).

Figure 3: Mechanistic biases of SV formation.

A) Cancer cells accumulate variants that are advantageous (positively selected drivers), or neutral or even deleterious (passengers). Even clones with neutral or deleterious events can expand if they coincide with drivers, or simply due to chance. B) The probability of DSB formation at a genomic locus depends on a combination of its characteristics, including proximity to active transcription (e.g. due to TOP2B-associated DSBs), repeat elements, GC content, and chromatin organization. C) The probability two loci fuse is primarily determined by their 3D distance. While this often correlates with genomic distance, DSBs in flexible regions can fuse to genomically distant loci due to 3D proximities. D) Following a DSB event multiple response factors including PARP1, ATM/ATR, and ALC1 are recruited to modify the surrounding histones. Collectively these changes result in a change of DNA topology towards increased mobility of the break ends.

The figures for this paper were produced by Nature Reviews Cancer graphics artists and are under copyright to Nature Reviews Cancer. The figures can be viewed at : Dubois F, Sidiropoulos N, Weischenfeldt J, Beroukhim R. Structural variations in cancer and the 3D genome. Nature Reviews Cancer. 2022 September;22(9):533–546. PMID: 35764888; DOI: 10.1038/s41568-022-00488-9.

The prevalence of an SV in cancer is thus determined both by the mechanistic biases that determine the probability with which the alteration occurs in the cancer’s cell of origin and by the extent to which effects on fitness in the context of evolutionary selection enrich the alteration in cells that reach cancer. The mechanistic biases shape the fraction of cells in a person’s body that acquire a given alteration. The effects on evolutionary fitness impact which of these cells will continue to grow, potentially forming a cancer. Likewise, once a cancer has formed, mechanistic biases continue to shape the fraction of subclones that generate any specific genetic alteration, whereas the fitness effects impact which of these subclones will grow to dominate the tumor.

Observing a genetic alteration at high frequency in cancer thus may suggest that the alteration was a driver event that increased cellular fitness, but does not provide proof. It is also possible that the alteration occurs frequently during cell division, including in cells that do not become cancer5,61,62.

4. Mechanistic biases shape where SVs form

a. SV formation understood on the linear genome

Mechanisms of SV formation fall into two major classes: replication-based mechanisms such as microhomology-mediated break-induced replication (MMBIR), in which a DNA polymerase jumps to a distant locus; and fusion-based mechanisms such as non-homologous end-joining (NHEJ), homologous recombination (HR), and microhomology-mediated end-joining (MMEJ), in which a double-strand break (DSB) in DNA is repaired incorrectly, fusing it to a distant locus.

Moreover, these mechanisms can generate single isolated SVs (Figure 1A) or multiple SVs, typically clustered in specific genomic loci, in a single traumatic event (Figure 1B). Such complex events often create DNA topologies that are difficult to interpret, with the potential for multiple distant loci brought into juxtaposition. These complex events can be generated by multiple mechanisms, including chromothripsis, breakage-fusion bridge cycles, extrachromosomal amplification, chromoanasynthesis, templated insertions, and chromoplexy⁶³.

Some of these complex events, such as chromothripsis, and the related formation of extrachromosomal amplifications and breakage-fusion-bridge events, can reflect DNA breaks that later fuse. In addition, complex events can form due to replication-based mechanisms. An example is chromoanasynthesis, which can form from repeated microhomology-mediated template switching and therefore is hypothesized to originate from MMBIR^64-66.

Many genomic features impact which loci break and fuse, including sequence characteristics and replication timing. However, the impact of the 3D structure of the genome on the development of rearrangements is becoming increasingly clear.

b. The genome’s 3D structure impacts where SVs form and fuse

In the context of SVs, mechanistic biases determine both which loci are prone to double-strand breakage and which pairs of loci are prone to combine.

The importance of 3D nuclear distance in cancer SV formation was identified in early, seminal observations. The fusion generating the Philadelphia chromosome⁶⁷, which joins BCR in chromosome 9 and ABL1 in chromosome 22 in myeloid leukemias, was demonstrated to be catalyzed at least in part by the 3D proximity of the two loci specifically in hematopoietic cells ^68,69. In contrast, these two loci are more distant in other cell types, where the fusion gene is not observed. Another well-established example is IGH-MYC, present in 9 out of 10 Burkitt’s lymphomas⁷⁰. However, a systematic, genome-wide assessment of the role played by the 3D organization was impossible until only recently.

The organization of the genome has impacts both on which loci form DSBs, and which pairs of loci join to form SVs. Indeed, DSBs are an intrinsic component of the maintenance of the 3D conformation of DNA during transcription. These DSBs are generated by the endonuclease topoisomerase 2B (TOP2B) to relieve tension on the double helix throughout the genome but especially adjacent to CTCF chromatin loop anchors (Figure 3B). Perhaps for this reason, sites of active transcription have been shown to be more prone to SV formation^5,71-74. These actively transcribed sites are also prone to selection, suggesting that the prevalence of SVs near open, active chromatin in cancer genomes may be a consequence of both mechanistic biases and their special effects on cellular fitness. In contrast, SNVs have been shown to primarily accumulate in late-replicating, repressed chromatin ⁷⁵.

Cell culture systems that trace DSBs through the cell cycle have shed light on processes that make these sites more prone to SVs. Untranscribed regions, heterochromatic chromatin loop anchors of the genome were found to be more often repaired during G1 through error-prone NHEJ, which is the most common source of SVs in cancer^20,76. In contrast, transcriptionally active sites of DSBs, associated with “A” compartment, long-range chromatin interactions, and histone marks of active transcription and HR-activity such as H3K79me2 and H3K36me3 were shown to be preferentially repaired by HR late during S-phase^77,78. Hence, DNA breaks at sites of active transcription can form clusters of long-range interactions in the nucleus, which are preferentially repaired during replication. This can be reconciled with a phase-separation model containing droplets with the accumulation of transcription factors, replication machinery, gH2AX, Ubiquitin, and 53BP1, as well as homologous recombination machinery. Indeed, 53BP1 forms liquid-like clusters around DNA breaks⁷⁹ and stabilizes the 3D conformation at sites of DSB by recruiting RIF1 to the TAD boundaries, which may serve to contain and concentrate DNA repair enzymes at sites of DSB⁸⁰. The choice of breakpoint fusion events during replication stress is also thought to be shaped by nuclear proximity⁸¹. Replication forks that are distant on the linear genome but close in space often exchange templates and form SVs ^82,83. Indeed, Hi-C based analysis of the 3D chromatin before and after induction of chromothripsis in a human cell line showed that SVs were significantly enriched between genomically distant loci with similar replication timing zones and compartments, in particular early replication timing and A compartments⁸⁴–which tend to be in close physical proximity^24,34,85.

Perhaps the most important factor influencing which pairs of DSBs fuse is the distance between them (the “span” of the SV). Across all SVs, the frequency with which rearrangements are observed decreases as the reciprocal of their spans. This distribution is the same as the distribution of distances between joined DNA fragments in Hi-C data–which has been interpreted to indicate a fractal globule three-dimensional structure of the genome ^5,20,23. This suggests that the three-dimensional conformation of DNA that determines which loci ligate in Hi-C experiments also determines which DSBs are likely to ligate in SVs in cancer, which corroborates findings from early high-resolution imaging ^70,86.

However, rearrangement spans are also influenced by local chromatin compaction (Fig. 3C). Once a DSB forms, its “search” for a site to ligate with appears to be enabled by the mobility of chromatin within the nucleus⁸⁷. DSBs can increase this mobility⁸⁸. Several chromatin changes occur upon DSB, which are thought to have a direct impact on the choice of DSB partner. Following HR, poly(ADP-ribose) polymerase 1 (PARP1) is recruited to the DSB and interacts with the chromatin remodeler ALC1 to relax the chromatin⁸⁹. ALC1 is frequently upregulated in cancer and is associated with resistance to PARP inhibitors⁹⁰. ATM and ATR kinases phosphorylate the histone variant H2AX (also known as gH2AX) at and around DSB sites. This occurs in conjunction with replacement of histone H1, widespread ubiquitination, and recruitment of the 53BP1 protein⁹¹. These chromatin changes affect DNA compaction and are thought to induce a change in the flexibility of the chromatin fiber, allowing for a more promiscuous search for partner locus (Fig. 3D). Indeed, sites with gH2AX were shown to partake in more chromatin interactions, both intra- and inter-chromosomal, and these sites were able to form clusters of chromatin⁹². Topologically associated domains are also known to influence somatic SV formation in the non-oncogenic process of VDJ recombination in developing B cells. Here, disruption of CTCF-mediated looping has been found to skew recombination ^93,94

Flexible genomic loci may also allow DSBs both to form more often and ligate with more distant loci. For example, common fragile sites (CFSs) are both prone to SV formation and tend to contain highly flexible regions. Other features enriched in these sites include AT-rich, G-negative chromosomal bands, ALU repeats, and late replication ⁹⁵. CFSs tend to involve late-replicating regions with highly transcribed large genes that span TAD boundaries. Indeed, by using polymer physics to describe interactions between chromatin fibers and chromatin-binding proteins without prior knowledge of nuclear proximity, it has been possible to predict many of the DNA interactions and 3D conformations observed in wild-type mouse cells and cells with acquired SVs from experimental Hi-C data ⁹⁶. Although this model was only applied to one well-studied 6 Mb locus associated with cancer-related SVs, it does suggest DNA folding shapes where SVs form. Therefore, the extent to which SVs occur in cancer genomes is to some extent dependent on the pre-cancerous tissue-specific 3D conformation^23,84,97,98. Although CTCF is associated with 75-95% of all TAD anchors ^41,99, adding CTCF binding sites to the model did not improve the correlation with experimental Hi-C data. Recently, however, a deep learning approach using Hi-C data found CTCF binding sites helped predict tissue and species-specific genome folding and SVs ¹⁰⁰.

In addition to topological tension, transcription-coupled processes can contribute to incomplete repair and SV formation (Figure 3B-C)^101,102. A prominent example is the fusion of the 5’ regulatory region of TMPRSS2 to the 2nd intron of the transcription factor proto-oncogene ERG, one of the most recurrent SVs in cancer. TMPRSS2 and ERG are located 2 Mb apart on chromosome 21 and chromatin conformation experiments have shown that these two genes are brought in close nuclear distance through chromatin looping in the prostate^5,103,104. TOP2B is recruited to relieve DNA tension by creating DSBs, resulting in TMPRSS2-ERG fusions due to illegitimate recombination between the 5’ end of TMPRSS2 and ERG. This leads to high-level upregulation of ERG through the activity of the TMPRSS2 promoter and enhancer. Curiously, in prostate cancer, SVs near regions associated with transcription factor binding, active transcription, and long-range chromatin looping, including those leading to TMPRSS2:ERG fusion, were shown to be highly enriched in younger patients⁷². The androgen receptor (AR) is a transcription factor that facilitates long-range interactions, and higher levels of AR in younger men might explain this observation. TMPRSS2:ERG formation has also been associated with chromoplexy formation involving one or several other chromosomes in the fusion formation⁸¹.

5. Fitness consequences of SVs

Most cancers are thought to evolve through a stepwise Darwinian process by sequential acquisition of somatic alterations, each associated with negative, neutral, or positive effects on cellular fitness. In contrast, the many rearrangements generated in a single complex SV can separately and simultaneously have positive and negative effects on fitness, providing a mechanism by which cancers undergo “punctuated evolution”^81,105-107. This is best exemplified by chromothripsis, hypothesised to be a stochastic process in terms of location and occurrence^108,109. Chromothripsis can cause several hundred SVs in a single event, including loss of multiple genomic loci. Essential genes—which comprise approximately a tenth of all genes¹¹⁰)—will often be disrupted by chance, resulting in strong negative selection. On rare occasions, some of these tens to hundreds of SVs can also have strong positive effects on fitness, resulting in increased overall fitness and survival of the tumour cells through this single catastrophic event^26,111.

While neutral selection is hypothesised to be common for SNVs and InDels (insertions or deletions of only a few bases)¹¹², it is considered uncommon for large SVs. These SVs typically have direct effects on a larger span of the genome--either through resulting large-scale changes in copy-number or by altering DNA topology and interactions--and therefore are more likely to affect one or more gene(s) either directly or indirectly^61,113,114.

The epigenetic landscape and 3D chromatin organisation of a cell have a large impact on the fitness effects of individual SVs. Although SVs can alter coding sequences and therefore gene composition, the major effects of SVs on selection are mediated through changes in gene dosage and in chromatin topology, both resulting in altered gene expression. These can be through trans (indirect) effects by affecting chromatin modifiers such as CTCF itself (discussed above). In multiple myeloma (MM), a translocation of the histone methyltransferase, NSD2 causes high-level upregulation, leading to pervasive methylation of H3K36. This was associated with changes in A/B compartment and TAD structure, which was linked with gene expression dysregulation in these tumors¹¹⁵. However, the main consequences of SV-mediated changes in 3D folding on selection rely on modifying direct interactions between cis-regulatory elements (CREs) and their target genes.

a. The importance of cis-regulatory elements

Transcriptional regulation is often retained within TADs, and early work demonstrated that disruption of TADs could lead to dramatic changes in the transcriptional regulation^9,116. Although one might expect that simply altering TAD boundaries would be the major mechanism by which SVs alter gene expression, this is usually not the case. According to a pan-cancer “PCAWG” survey across 2,700 tumor genomes, SVs and especially duplications often span TAD boundaries¹¹⁷ However, only a minority of these TAD-disrupting SVs were associated with marked changes in gene expression. This is in line with experiments showing minimal effect on gene expression following reduction of TADs through CTCF or Cohesin depletion^99,118. In contrast, both the availability of open chromatin, enabling transcription factor binding, and the specific locus constituents are important factors in shaping the effects of SVs on transcription. However, the exact constituents mediating SV-associated mis-expression is a major open question.

Two important constituents appear to be ‘active’ promoters and enhancers that can translate transcription factor binding into altered expression. Both of these are considered to be cis-regulatory elements (CREs).: Promoters broadly serve as a landing platform to recruit transcription factors and RNA Polymerase II to the transcription start site (TSS). The classical model of gene activation involves long-range enhancer-promoter contact (abbreviated E:P from hereon) ¹¹⁹. Enhancers have operationally been categorised into latent, primed, and active state¹²⁰, based on their ability to stimulate transcription, and are typically found to bind tissue-specific TFs and form long-range E:P looping. Promoters and enhancers are therefore classically thought to be involved in transcription initiation and transcription amplification, respectively, although these distinctions do not always hold and they can in some instances substitute for one another (see recent reviews^121,122). Enhancers can form long-range contact with one or more promoters, depending upon their nuclear distance and the presence of cofactors such as transcription factors. By altering the distances between these regulatory elements, SVs can generate new contacts, which may result in SV-mediated oncogenic activation in cis. Experiments in engineered mouse embryos found that only SVs that directly ‘flipped’ active enhancers towards promoters caused significant expression changes¹²³, whereas simply deleting CTCF binding sites had minimal effects--findings that are consistent with the results from the PCAWG analysis of cancer genomes. A key requirement for SV-mediated gene expression upregulation in cis therefore appears to be de novo loop formation between CRE, and indeed the majority of SV-mediated gene expression upregulation in cis appears to involve gene-proximal promoter regions being placed in closer proximity to active, normally distal regulatory regions associated with enhancer activity⁵, a mechanism coined enhancer hijacking⁹.

These CREs, and especially enhancers, often display considerable variation across species, tissues, and differentiation stages ¹²⁴. Perhaps for this reason, oncogenic fusion events that exploit enhancer hijacking tend to be highly tissue- or even cell-type-specific ¹²⁵ The exact enhancer altered to activate the proto-oncogene MYC, for example, is highly dependent on the tumor and tissue type and involves medulloblastoma, B-cell, T-cell, myeloid, endometrial, lung and colorectal cancer-specific enhancers^96,103,104.

b. Mechanisms by which SVs modify interactions between genes and cis-regulatory elements

Although the exact details remain to be identified, several mechanisms have emerged by which CRE-gene interactions are modified (Figure 4A). We have also made a website platform to visualise SVs and TADs across 22 tumor types.

Figure 4: Effects on fitness that shape SVs in cancer.

A) Five mechanisms by which SVs can increase expression of nearby oncogenes. (1) Enhancer juxtaposition. A lineage-specific enhancer, which promotes tissue-specific expression, is juxtaposed with an oncogene. (2) CRE-gene fusions. Additional cis-regulatory elements such as promoters are juxtaposed with the oncogene. (3) Enhancer de novo looping. The SV generates new long-range loops involving lineage-specific enhancers. (4) Enhancer amplification. A low-activity enhancer is amplified, multiplying its effects on expression of oncogenes within its TAD. (5) Extrachromosomal amplification. Combinations of the three prior mechanisms are active in extrachromosomal amplicons. These often contain enhancers hijacked from distant sites, and their circular topology circumvents insulators, allowing enhancers from the whole amplicon to interact with contained oncogenes. B) Positions of oncogenic SVs and their mechanisms of activation, indicated by the colored circles below affected oncogenes.

The figures for this paper were produced by Nature Reviews Cancer graphics artists and are under copyright to Nature Reviews Cancer. The figures can be viewed at : Dubois F, Sidiropoulos N, Weischenfeldt J, Beroukhim R. Structural variations in cancer and the 3D genome. Nature Reviews Cancer. 2022 September;22(9):533–546. PMID: 35764888; DOI: 10.1038/s41568-022-00488-9.

SV-mediated Enhancer Juxtaposition.

The seminal study introducing the concept of enhancer hijacking detected a set of highly recurrent SVs on chromosomes 1 and 9 in pediatric group 3 and 4 medulloblastomas. By integrating SVs with chromatin data and mRNA expression, the authors identified GFI1 and GFI1B as highly upregulated due to juxtaposition of distant, active enhancers⁹. Since then, similar examples of enhancer hijacking have been observed across multiple SVs and cancer types^7,126-128.

One of the oncogenes that are most often activated by this mechanism is TERT, the catalytic enzyme of telomerase. Almost all cancers require telomerase activation to prevent telomere attrition and induce immortalisation. TERT promoter-associated recurrent SVs were first found to be associated with high-level upregulation in lymphoma¹²⁹ and later across different cancer types^{130,131,7,21,132}. Mapping of the distal breakpoints has identified a wide range of partner loci, which tend to be associated with open, active chromatin.

Although SV-mediated enhancer hijacking primarily involves enhancers from cancer-irrelevant genes, an example of ‘enhancer swapping’ between the two oncogenes BCL6 and MYC was observed in B-cell lymphoma. Curiously, other B-cell lymphomas involved enhancer amplification of the BCL6 super-enhancer, leading to its upregulation ¹³³.

Although whether TADs are a functional unit is under debate, chromatin features associated with TAD boundaries likely prevent illegitimate enhancer:promoter (E:P) interactions and de novo TAD formation can create novel E:P interactions. A prominent example is in colorectal cancers, where a series of highly recurrent tandem duplications spanning both the oncogene IGF2 and a strong, colon-specific enhancer in a neighbouring TAD was found to cause formation of a so-called neo-TAD, leading to more than 200-fold upregulation of IGF2 ⁷.

SV-mediated CRE-gene Fusion.

In some cases, other CRE elements such as promoters are also hijacked along with the enhancers by the fusion event. TMPRSS2:ERG is an example of SV-mediated CRE-gene fusion, whereby the entire regulatory machinery of TMPRSS2 is recruited to ERG to drive its expression. Of note, in addition to driving high-level upregulation of ERG, the fusion also removes a degron motif in the N-terminus of ERG, making it resistant to ubiquitin-mediated degradation ^134,135. To date, we know of no cases in which CRE elements are recurrently hijacked without including enhancers.

In some cases, single SVs are oncogenic through two or more such mechanisms. For example, inversions in chromosome 3 in leukemia activate the protooncogene MECOM by hijacking enhancers of the haploinsufficient tumor suppressor gene GATA2, and thereby simultaneously reduce GATA2 expression ¹³⁶. Another intriguing example is the intrachromosomal deletion leading to fusion between QKI and MYB on chromosome 6 and concomitant upregulation of MYB, truncation of its negative regulatory domain, and partial loss of the tumor suppressor QKI ^8,137.

In a minority of cases, these types of fusions can also generate proteins with de novo functions ⁵, which are the subject of other reviews¹³⁸

SV-mediated Enhancer Looping.

Although the majority of SVs causing TAD disruption do not alter gene expression, certain events have been associated with de novo long-range looping and oncogenic dysregulation of cancer genes. Indeed, CTCF binding sites represent some of the most frequently mutated non-coding sites in the cancer genome ^139,140. These mutations can disrupt CTCF-mediated insulator effects, generate novel E:P long-range loops, and activate oncogenes ^141,142. A systematic pan-cancer analysis of recurrent SVs that disrupt TADs identified several instances of oncogene activation through altering the 3D chromatin architecture ⁷. The most prominent example was a set of highly recurrent SVs in lung squamous carcinoma that intersected a CTCF loop-anchor boundary and led to high-level upregulation of the oncogene IRS4, situated 100 kb away. SV-mediated disruption of the TAD boundary led to spreading of active chromatin, marked by H3K27ac, and ectopic looping between a normally silent enhancer and the IRS4 promoter. Similar findings have been observed for cancers with mutations or deletions in CTCF binding sites, which can lead to permissive, long-range looping ^143,144.

Many recurrent “hot spots” of SVs with unknown gene targets may indeed be explained by formation of de novo long-range enhancer-promoter contacts, but the distances can be several hundred kb. A prototypical example of SV mediated long-range looping was identified in medulloblastoma, where a series of complex SVs including tandem duplications and an inversion, all spanning a super-enhancer and a TAD boundary, caused the enhancer to ‘flip’ to the other side of the TAD boundary, leading to high-level upregulation of the histone methyltransferase PRDM6, residing more than 600kb away from the enhancer ¹²⁶.

SV-mediated Enhancer amplification.

Copy number amplifications can amplify whole genes to cause gene expression dosage increase. However, amplification of strong, lineage-specific enhancers that control MYC have been identified in AML, T-ALL, lung, and endometrial carcinoma ^145,146. By analysing recurrent amplified non-protein-coding regions in the genome, Zhang et al conducted a systematic analysis of recurrently amplified non-protein-coding genomic loci and identified recurrent amplifications of four regulatory regions that were associated with increased expression of otherwise unaltered genes controlled by these enhancers, including MYC ¹⁴⁷. Similar to the enhancer hijacking examples, the enhancer amplifications were distinct and specific to each tumor type, again arguing for a strongly tissue-specific epigenetic mechanism.

A similar mechanism of enhancer amplification was identified in prostate cancer, where an androgen receptor-specific lineage-specific enhancer was found to be frequently amplified in response to androgen-deprivation, leading to re-activation of androgen receptor expression through increased E:P interaction ¹⁴⁸.

Enhancer Looping in Extrachromosomal DNA.

Extrachromosomal circular DNA (ecDNA) structures can be present in hundreds of copies and have often been found to carry one or more oncogenes ^25,149,150. The upregulation of ecDNA-containing oncogenes are thought to be mediated primarily through dosage-effects, but recent discoveries suggest that increased chromatin accessibility in ecDNAs ²⁷, as well as noncoding elements associated with enhancer activity from other regions of the genome, can be incorporated into the ecDNA, possibly causing novel E:P interactions with the oncogenes. Examples include ecDNA incorporating EGFR and MYC and an enhancer from an adjacent TAD in glioblastoma and neuroblastoma ^15,151. Recent discoveries also showed that ecDNA can act in trans as enhancers for promoters on the linear genome¹⁵² with other ecDNA molecules in a structure called the ecDNA hub, which can facilitate co-evolution and synergies between amplicons¹⁵³.

5. Conclusion

Much of cancer genome discovery to date has focused on the exome. However, SVs by nature tend to occur within the 99% of the genome that does not encode proteins, and which is the domain of gene regulatory elements. As technologies improve to characterize both SVs genome-wide and the characteristics of the regulatory elements across the genome, we are learning more about the essentiality of their interaction. Going forward, we anticipate that a relatively complete understanding of the specific epigenetic contexts within which each SV occurs will be essential to understanding both the mechanisms by which those SVs are formed, their effects on cancer evolution, and their clinical interpretation. An important consideration when linking distant SVs to gene expression in cis, such as enhancer hijacking, is to phase the SVs with expression data to the same allele, which is currently challenged by short read length and short haplotype blocks (see Box 1).

For this reason, long-read sequencing will likely help resolve both the genome topology resulting from SVs and their relations to epigenetic elements that impact chromatin conformation. Several additional technologies may also help shed light on the causal relations between 3D genome structure and SV formation. For example, single-cell DNA sequencing can be used to identify the mechanistic biases associated with SV formation before cell division and selection, by detecting DNA breaks and rearrangement formation at single-cell resolution before each cell undergoes cell division⁵⁷. Complete telomere-to-telomere¹⁵⁴ and ancestry-aware human reference assemblies will also enable a more accurate and sensitive mapping and analysis of changes in DNA topology. Moreover, large-scale profiling of chromatin conformations across cell and tissue types may also support an improved understanding of the relations between tissue-specific DNA organization and SV rates. Similarly, sequential analyses of genomes during cancer formation and progression may support a more detailed understanding of how SVs form in loci with altered 3D organization, such as extrachromosomal DNA. Lastly, much larger numbers of cancers will have to undergo whole genome sequencing to enable robust detection of SVs that recur more frequently than current models would suggest⁵.

As our knowledge and tools advance, we anticipate additional questions to emerge. Perhaps the most prominent and immediate of these is a full understanding of chromatin interactions in 3D and how these relate to somatic SVs. Although Hi-C and related technologies have been available for several years now, their high costs in both money and DNA have severely limited the contexts in which they have been applied. Recent discoveries have also highlighted the promiscuity of certain promoters with varied enhancers (e.g. MYC), whereas other promoter-enhancer interactions appear to be more restrictive. The reasons behind the selectivity of promoter-enhancer interactions are unclear.

Perhaps in the more distant future, we anticipate contextual questions to become more prominent. For example, the epigenetic contexts in which SVs occur evolve with cellular differentiation, including within tumors. How do these changes in epigenetic context within a tumor modify the effects of the SVs in the tumor, and do these SV-epigenetic interactions, in turn, limit the cellular differentiation states within the tumor? Cancer cells tend to reside in progenitor-like cell states; does this create a common and unifying stem-like epigenetic background across cancers? Moreover, the vast majority of germline genomic variation occurs outside the exome, and its effects on phenotype are thought to occur largely through modifications of CREs. How do these germline variants affect SV formation and selective effects, and do these relate to cancer risk both within and across ancestral populations?

Lastly, how do we ultimately leverage this understanding of the role of epigenetics in SV formation and selective effects to develop new cancer preventative and therapeutic approaches? Existing therapeutics that target oncogenes activated by SVs, such as fusions of BCR-ABL1 and PML-RARα, and amplifications of ERBB2 and EGFR, have targeted the enzymatic activity of the oncogenes themselves. However, an improved understanding of changes in the 3D genome that occur with SVs may offer new opportunities. Altered interactions between oncogenes and cis-regulatory domains will not alter the amino acid sequence of the oncogene itself, but only the transcriptional regulation; epigenetically-directed therapies (reviewed in ^155,156) may have high specificity in these cases. For example, a recent study demonstrated that ecDNA- containing cancers were susceptible to the BET domain inhibitor JQ1, due to the high, focal concentration of transcriptional activity at ecDNA hubs¹⁵³. In some cases, detection of SVs altering 3D chromatin structure may serve important diagnostic purposes. For example, the androgen receptor is often activated by amplification of its enhancer that do not include the gene itself^148,157. Diagnostic and therapeutic approaches that can effectively detect and leverage the altered epigenetic interactions induced by SVs in cancer cells would be likely to have a major impact on patient outcomes.

Glossary:

Branching evolution

A pattern of tumor evolution in which subclones with distinct driver combinations emerge from the initial clonal outgrowth of the tumor, contributing to tumor heterogeneity

Breakage-Fusion Bridge cycles

A mechanism of amplification in cancer genomes. Chromosomes fuse due to telomere shortening and get broken apart during mitosis leaving one daughter cell with extra copies of genes.

Chromoanasynthesis

A mechanism of complex SV generation due to template switching during replication

Chromoplexy

Chains of SVs involving multiple chromosomes

Chromatin compartments

Subdivisions of chromosome territories divided into compartments with open and closed chromatin

Chromosome Territories

A discrete region within the nucleus containing a specific chromosome

Chromothripsis

A mechanism of SV generation leading to a complex pattern of step-like copy number loss and random relegation of DNA fragments resulting in approximately equal ratios of deletion to duplication and the two inversion types.

Driver events

Genetic variants resulting in increased evolutionary fitness of the affected cell.

Evolutionary Selection

The process of enrichment of clones with higher fitness in a population of cancer cells.

Extrachromosomal Amplification

A mechanism of amplification in cancer genomes potentially originating as a byproduct of chromothripsis. Extrachromosomal amplicons are small circular fragments of DNA that typically contain at least one oncogene but no centromere. Due to the lacking centromere, they are randomly distributed during mitosis. The daughter cell receiving more amplicons can gain evolutionary fitness leading to rapid high-level amplification of the affected oncogenes in the cancer.

Hi-C

A proximity ligation-based sequencing technology that, in principle, allows the detection of all 3D interactions between DNA segments in the nucleus.

MMBIR

Microhomology-Mediated Break-Induced Replication, a form of microhomology-mediated template switching during replication, which can lead to a complex SV pattern.

Passenger events

Genetic variants with no effect on the evolutionary fitness of the affected cell. Phase separation: A phenomenon of decreased mixing between molecules because of differences in the intermolecular interactions. Most familiar from the unmixing of oil and water resulting in two separate phases.

Punctuated evolution

Bursts of changes to the genome, with many variants acquired in one event, resulting in a dramatically changed evolutionary fitness. Often contrasted with the sequential acquisition and selection of individual variants in classic evolutionary theory.

Selective sweep

Outgrowth of a subclone with increased fitness that completely replaces the parental and all other clones in the population.

Templated Insertions

A form of SV caused by the DNA polymerase jumping to a different strand during replication.

Topologically associated domains (TADs)

Regions in the genome in the range of several 100kbps - few Mbp which are separated by boundary elements and show higher interaction frequencies in 3D within one TAD than between TADs.