Enhancer hijacking determines intra- and extrachromosomal circular MYCN amplicon architecture in neuroblastoma

MYCN amplification drives one in six cases of neuroblastoma. The supernumerary gene copies are commonly found on highly rearranged, extrachromosomal circular DNA. The exact amplicon structure has not been described thus far and the functional relevance of its rearrangements is unknown. Here, we analyzed the MYCN amplicon structure and its chromatin landscape. This revealed two distinct classes of amplicons which explain the regulatory requirements for MYCN overexpression. The first class always co-amplified a proximal enhancer driven by the noradrenergic core regulatory circuit (CRC). The second class of MYCN amplicons was characterized by high structural complexity, lacked key local enhancers, and instead contained distal chromosomal fragments, which harbored CRC-driven enhancers. Thus, ectopic enhancer hijacking can compensate for the loss of local gene regulatory elements and explains a large component of the structural diversity observed in MYCN amplification.


Introduction
Oncogene amplification is a hallmark of cancer genomes. It leads to excessive proto-oncogene overexpression and is a key driver of oncogenesis. The supernumerary gene copies come in two forms, i. self-repeating arrays on a chromosome (homogeneously staining regions, HSR) and ii. many individual circular DNA molecules (extrachromosomal DNA, ecDNA, alias double minute chromosomes, dmin) 1 . EcDNA can arise during genome reshuffling events like chromothripsis and are subsequently amplified 2,3 . This partially explains why such circular DNAs can consist of several coding and non-coding distant parts of one or more chromosomes 4 .
Over time, amplified DNA acquires additional internal rearrangements as well as coding mutations, which can confer adaptive advantages such as resistance to targeted therapy 5-7 .
EcDNA re-integration into chromosomes can lead to intrachromosomal amplification as HSRs 8,9 and act as a general driver of genome remodeling 10 . Our knowledge of the functional relevance of non-coding regions co-amplified on ecDNA, however, is currently limited.
MYCN amplification is a prototypical example of a cancer-driving amplification. The developmental transcription factor was identified as the most commonly amplified gene in a recent pediatric pan-cancer study 11 . Its most prominent role is in neuroblastoma, a pediatric malignancy of the sympathetic nervous system. MYCN amplification characterizes one in six cases and confers dismal prognosis 12 . In contrast to long-term survival of more than 80% for non-amplified cases, 5-year overall survival is as low as 32% for MYCN-amplified neuroblastoma 12 . In these cases, MYCN amplification is likely an early driver of neuroblastoma formation. Accordingly, MYCN overexpression is sufficient to induce neuroblastic tumor formation in mice 13,14 . Despite its central role in neuroblastoma biology, the epigenetic regulation of MYCN is incompletely understood.
Recently, studies have identified a core regulatory circuit (CRC) including half a dozen transcription factors that drive a subset of neuroblastoma with noradrenergic cell identity, including most MYCN-amplified cases [15][16][17][18] . The epigenetic landscape around MYCN is less well described. In part, this is due to the structural complexity of MYCN amplicons and difficulties in the interpretation of epigenomic data in the presence of copy number variation. Recent evidence has emerged suggesting that local enhancers may be required for proto-oncogene expression on amplicons 19 . Here, we sought out to identify key regulatory elements near MYCN in neuroblastoma by integrating short-and long-read genomic and epigenomic data from neuroblastoma cell lines and primary tumors. We investigated the activity of regulatory elements in the context of MYCN amplification and characterized the relationship between amplicon structure and epigenetic regulation.

Local CRC-driven enhancers contribute to MYCN expression in neuroblastoma
In order to identify candidate regulatory elements near MYCN, we examined public H3K27ac chromatin immunoprecipitation and sequencing (ChIP-seq) and RNA sequencing (RNA-seq) data from 25 neuroblastoma cell lines 15 . ChIP-seq data for amplified genomic regions are characterized by a very low signal-to-noise ratio, which has complicated their interpretation in the past 16 . We therefore focused our analysis on 12 cell lines lacking MYCN-amplifications but expressing MYCN at different levels, allowing for the identification of MYCN-driving enhancers in neuroblastoma. Comparison of composite H3K27ac signals of MYCN-expressing vs. non-expressing cell lines identified at least 5 putative enhancer elements (e1-e5) that were exclusively present in the vicinity of MYCN in cells expressing MYCN, thus likely contributing to MYCN regulation (Fig. 1, Supplementary Fig. 1a). Consistent with differential RNA expression, a strong differential H3K27ac peak was identified spanning the MYCN promotor and gene body (MYCNp; Fig. 1). The identified enhancers were not active in developmental precursor cells such as embryonic stem cells, neuroectodermal cells or neural crest cells ( Supplementary Fig. 1b), suggesting these enhancers were specific for later stages of sympathetic nervous system development or neuroblastoma. Transcription factor ChIP-seq in MYCN-expressing cells confirmed that four of the enhancers (e1, e2, e4, e5) were bound by each of three noradrenergic neuroblastoma core regulatory circuit factors (PHOX2B, HAND2, GATA3; Fig. 1b). All but enhancer e3 harbored binding motifs for the remaining members of the core regulatory circuit (ISL1, TBX2, ASCL1; Supplementary Fig. 1c) for which ChIP-seq data were unavailable. Additionally, all enhancers contained binding motifs for TEAD4, a transcription factor implicated in a positive feedback loop with MYCN in MYCN-amplified neuroblastoma 20 . Two of the enhancers (e1, e2) also harbored canonical E-boxes, suggesting binding of MYCN at its own enhancers ( Supplementary Fig. 1c). Thus, a common set of CRCdriven enhancers is found specifically in MYCN expressing neuroblastoma cells, indicating that MYCN expression is regulated by the CRC.

Local enhancer co-amplification explains asymmetric MYCN amplicon distribution
MYCN is expressed at the highest levels in neuroblastomas with MYCN amplifications ( Supplementary Fig. 1d). It is unclear, however, to what extent enhancers are required for sustained MYCN expression on MYCN-containing amplicons. To address this, we mapped amplified genomic regions in a meta-dataset of copy-number variation in 240 MYCN-amplified neuroblastomas 21 . This revealed an asymmetric pattern of MYCN amplification (Fig. 2a, Supplementary Fig. 2). Intriguingly, a 290kb region downstream of MYCN was co-amplified in more than 90% of neuroblastomas, suggesting that MYCN amplicon boundaries were not randomly distributed, which is in line with recent reports in a smaller tumor cohort 19 Notably, the consensus amplicon boundaries did not overlap with common fragile sites ( Supplementary   Fig. 2g), challenging a previous association found in ten neuroblastoma cell lines 8 . Regions of increased chromosomal instability alone are therefore unlikely to explain amplicon boundaries.
Intriguingly, several MYCN-specific enhancers were found to be commonly co-amplified (Fig.   2b). The distal MYCN-specific CRC-driven enhancer, e4, was part of the consensus amplicon region in 90% of cases. Randomizing amplicon boundaries around MYCN showed that e4 coamplification was significantly enriched on MYCN amplicons (empirical P=0.0003). Coamplification frequency quickly dropped downstream of e4, suggesting that MYCN-specific, CRC-driven enhancers are a determinant of MYCN amplicon structure and may be required for MYCN expression, even in the context of high-level amplification.

Distal CRC-driven super enhancers are significantly co-amplified with MYCN in neuroblastoma
We and others have previously described chimeric MYCN amplicons 10 containing distal chromosomal fragments. We therefore systematically inspected MYCN-distal regions on chromosome 2 for signs of co-amplification. Distinct regions were statistically enriched for coamplification with MYCN (Fig. 2c). In line with previous reports 22 , significant co-amplification of 19 protein-coding genes, including known neuroblastoma drivers such as ODC1, GREB1 and ALK occurred in MYCN-amplified neuroblastoma. Intriguingly, co-amplification of distal CRC-driven super enhancers (SE) occurred in 23.3% of samples. Seven specific CRC-driven SEs were significantly co-amplified more often than expected by chance. Most of these SEs were found in gene-rich regions, precluding to determine whether genes or regulatory elements were driving co-amplification. One significantly co-amplified CRC-driven SE, however, was found in a gene-poor region in 2p25.2, where most co-amplified segments did not overlap protein-coding genes (Fig. 2c). This raised the question whether hijacking of such distal regulatory elements may explain co-amplification with MYCN.

Enhancers remain functional on MYCN amplicons
Based on our amplicon boundary analysis, two classes of MYCN amplicons could be distinguished in neuroblastoma, i. amplicons containing local MYCN-specific enhancers, including e4, (here referred to as class I amplicons; Fig. 3a) and ii. amplicons lacking local MYCN-specific enhancers, and at least lacking e4 (referred to as class II amplicons; Fig. 3b).
To determine whether co-amplified enhancers were active, we acquired genomic (long-and short-read whole genome sequencing) and epigenomic (ATAC-seq and H3K4me1 and H3K27ac ChIP-seq) data for two neuroblastoma cell lines with class I amplicons (Kelly and NGP) and two neuroblastoma cell lines with class II amplicons (IMR-5/75 and CHP-212).
Notably, H3K27ac signal-to-noise ratio was lower on MYCN amplicons than in non-amplified regions. While the fraction of reads in peaks on the amplicon did not clearly differ between the amplicon and randomly drawn genomic regions, we observed more peaks than for nonamplified regions ( Supplementary Fig. 3). These peaks were characterized by a lower relative signal compared to the amplicon background signal, indicating a larger variety of active regulatory regions on different MYCN amplicons. Using Nanopore long read-based de novo assembly, we reconstructed the MYCN neighborhood, confirming that MYCN and e4 were not only co-amplified in class I amplicons, but also lacked large rearrangements, which could preclude enhancer-promoter interaction ( Supplementary Fig. 4-5). Enhancer e4 was characterized by increased chromatin accessibility and active enhancer histone marks as determined by ATAC-seq, H3K4me1 and H3K27ac ChIP-seq (Fig. 3c). Importantly, 4C chromatin conformation capture analysis showed that e4 spatially interacted with the MYCN promotor on the amplicon (Fig. 3c). Thus, e4 presents as a functional enhancer and appears to contribute to MYCN expression even in the context of class I MYCN amplification.

Super enhancer hijacking compensates for the loss of local enhancers on chimeric intra-and extrachromosomal circular MYCN amplicons
In contrast to class I amplicons, class II amplicons did not include local enhancers, raising the possibility of alternative routes of MYCN regulation. The lack of a strong local regulatory element on class II amplicons and our observation of frequent co-amplification of distal SE ( Fig. 2c), led us to hypothesize that ectopic enhancers might be recruited to enable MYCN expression in class II amplicons. In line with our hypothesis, primary neuroblastomas with class II amplicons were more likely to harbor complex amplifications containing more than one fragment (66.7% vs. 35.7%, Fisher's Exact Test P=0.003; Fig. 3e). All class II amplicons coamplified at least one CRC-driven super enhancer element distal of MYCN. Some enhancers were recurrently found on class II amplicons, including an enhancer 1.2Mb downstream of MYCN that was co-amplified in 20.8% (5/24) of MYCN-amplified neuroblastomas, 2.1-fold higher than expected for randomized amplicons that include MYCN but not e4 (Fig. 3f). Thus, class II MYCN amplicons are of high chimeric structural complexity allowing for the replacement of local enhancers through hijacking of distal CRC-driven enhancers.
To determine the structure and epigenetic regulation of class II amplicons in detail, we To analyze the interaction profile in circular and linear amplicons we performed Hi-C and mapped the reads to the reconstructed amplicon ( Fig. 4c, g). This analysis supported the genomic sequencing-based reconstruction of the amplicon, recapitulating the order and orientation of the joined fragments and confirmed that the ectopic enhancers spatially interacted with MYCN. Notably, high-frequency interactions in the corners of the maps opposite to the main diagonal, confirmed the circularity of CHP-212 amplicon and the presence of tandem amplification in IMR-5/75. In IMR-5/75 and CHP-212, we observed insulated TADs, boundaries and loops as in the rest of the genome. Due to the rearrangements in CHP-212, the MYCN gene became part of a neo-TAD consisting of a sub-TAD that originated from the wild type genome as an intact unit, and a second sub-TAD that resulted from the fusion and coamplification of the first region with another region from a distal part of chromosome 2 (chr2:12.6-12.8Mb), containing multiple CRC-driven SEs (Fig. 4g, Supplementary Fig. 6b).

Nanopore long-read DNA sequencing can be used for parallel assessment of MYCN amplicon structure and epigenetic regulation
In addition to allowing the alignment-free de novo assembly of the MYCN amplicon in several samples ( Fig. 4b-d, f-h, Supplementary Fig. 4-5), Nanopore sequencing also allows for the direct measurement of DNA methylation without the need for bisulfite conversion (Fig. 5a) 23 .
While DNA methylation at regulatory elements is often associated with repression, a trough in DNA methylation may indicate a transcription factor binding event, a poised or active gene regulatory element, or a CTCF-occupied insulator element (Fig. 5b). In theory, Nanopore sequencing and assembly might allow for the simultaneous inference of both structure and regulatory landscape (Fig. 5b). Prior to evaluating the MYCN amplicons, the DNA methylation landscape of highly expressed and inactive genes demonstrated the expected distribution of decreased methylation at active promoters and increased methylation within active gene bodies ( Fig. 5c). In order to assess the DNA methylation status of putative regulatory elements near MYCN, we first used the amplicon-enriched ATAC-seq peaks to classify relevant motif signatures (Fig. 5d). While MYCN was surrounded by the expected CRC-driven regulatory elements at the overlapping core enhancers as well as some CTCF sites, both their number and location varied, indicative of sample-specific sites of regulation. Indeed, DNA methylation decreased in accordance with sites specific to a given sample (Fig. 5e), opening up the possibility of using these data to infer regulatory elements in patient samples when no orthogonal epigenomic data are available.

Class II MYCN amplicons clinically phenocopy class I amplicons
MYCN-amplified neuroblastoma is characterized by significant clinical heterogeneity, which cannot entirely be explained genetically. Whether the structure of the MYCN amplicon itself could account for some of this variation is currently unknown. In line with previous reports 22 , higher counts of amplified fragments were associated with a more malignant clinical phenotype ( Fig. 6a). Co-amplification of ODC1, a gene located 5.5Mb upstream of MYCN and coamplified in 9% (21/240) of MYCN-amplified neuroblastomas (Fig. 2c), defined an ultra-high risk genetical subgroup of MYCN-amplified neuroblastoma (HR 2.3 (1.4-3.7), Log-rank test P=0.001; Fig. 6b). Similarly, ALK co-amplification, present in in 5% (12/240) of MYCNamplified tumors, was also associated with adverse clinical outcome (HR 1.8 (0.94-3.4), Logrank test P=0.073; Fig. 6c). In contrast, differences in the MYCN amplicon enhancer structure, i.e. class II amplification, did not confer prognostic differences (HR 1.3 (0.78-2.1), Log rank test P=0.34; Fig. 6d). We therefore conclude that chimeric co-amplification of proto-oncogenes partly explain the malignant phenotype of neuroblastomas with complex MYCN amplicons, whereas enhancer hijacking in class II amplicons does not change clinical behavior, fully phenocopying class I MYCN amplicons.

Discussion
Here, we show that neuroblastoma-specific CRC-driven enhancers contribute to MYCN amplicon structure in neuroblastoma and retain the classic features of active enhancers after genomic amplification. While most MYCN amplicons contain local enhancers, ectopic enhancers are regularly incorporated into chimeric amplicons lacking local enhancers, leading to enhancer hijacking.
A large subset of neuroblastomas was recently found to be driven by a small set of transcription factors that form a self-sustaining core regulatory circuit, defined by their high expression and presence of super-enhancers [15][16][17][18] . In how far MYCN itself is directly regulated by CRC factors was previously unclear, particularly due to the challenging interpretation of epigenomic data on amplicons 16 . Our results provide empiric evidence that MYCN is driven by CRC factors even in the context of MYCN amplification. This is in line with and can mechanistically explain the previous observation that genetic depletion of CRC factors represses MYCN expression even in MYCN-amplified cells 16 . The finding that ectopic enhancers driven by the CRC are juxtaposed to MYCN on amplicons that lack local enhancers further strengthens the relevance of the CRC in MYCN regulation.
In line with our observation of local enhancer co-amplification, Morton et al. recently described that local enhancers are significantly co-amplified with other proto-oncogenes in other cancer entities 19 . They showed that experimentally interfering with local EGFR enhancers in EGFR- Reconstruction of amplicons has previously relied on combining structural breakpoint coordinates to infer the underlying structure. This regularly resulted in ambiguous amplicon reconstructions, which had to be addressed by secondary data such as Chromium linked reads or optical mapping 4,6,24 . We demonstrate the feasibility of long-read de novo assembly for the reconstruction of amplified genomic neighborhoods. De novo assembly was able to reconstruct entire ecDNA molecules and confirm the tandem duplicating nature of homogeneously staining regions. Integrating de novo assembly with methylation data from Nanopore sequencing reads will likely benefit further studies of other proto-oncogene-containing amplicons by enabling the characterization of the interplay between structure and regulation in highly rearranged cancer genomes.
Functional studies have shown that both ODC1 and ALK are highly relevant in neuroblastoma 28,29 . Co-amplification with MYCN has been reported before 22 , but to our knowledge the clinical relevance of co-amplification had not been determined so far. Similar to our previous observations of PTP4A2 co-amplification on chimeric ecDNA 10 , we demonstrate here that proto-oncogenes reside side-by-side on the same extrachromosomal circular DNAs, sometimes even sharing the same regulatory neighborhood. It is tempting to speculate that this structural coupling of genes could confer MYCN-independent but MYCN-amplicon-specific, collateral therapeutic vulnerabilities in MYCN-amplified tumors.
We conclude that the structure of genomic amplifications can be explained by selective pressure not only on oncogenic coding elements, but also on non-coding regulatory elements. CRCdriven enhancers are required for successful MYCN amplification and remain functional throughout this process. Even though the majority of amplicons contain endogenous enhancers, these can be replaced by ectopic CRC-driven elements that are juxtaposed to the oncogene through complex chimeric amplicon formation. We envision that our findings also extend to oncogene amplifications in other cancers and will help identify functionally relevant loci amongst the diverse array of complex aberrations that drive cancer.

ChIP-seq
As reported before 27 , cells were digested with Trypsin-EDTA 0.05% (Gibco) for 10 min at 37 °C. The cells were mixed with 10% FCS-PBS, and a single-cell suspension was obtained using a 40-µm cell strainer 30

ATAC-seq
ATAC-seq samples were processed as reported in Buenrostro et al 32 . 5x10 5 cells were used per sample. For sequencing, libraries were generated using Illumina/Nextera adapters and size selected (100-1000bp) with AMPure Beads (Beckman Coulter). Approximately 100 million 75bp paired-end reads were acquired per sample on the HiSeq 2500 system (Illumina).
Additional public ATAC-seq FASTQ files were downloaded from Gene Expression Omnibus (GSE80154) 33 . Adapter trimming, alignment and duplicate removal as for ChIP-seq. We generated BigWig tracks by extending paired-end reads to fragment size, filtering by the ENCODE DAC blacklist and normalizing to counts per million in 10bp bins (Deeptools 3.3.0).
Peaks were called using MACS2 (2.1.2) with default parameters.

Hi-C
3C libraries for Hi-C and 4C were prepared from confluent neuroblastoma cells according to the cell culture section above. Hi-C experiments were performed as duplicates. Cells were washed twice with PBS and digested with Trypsin-EDTA 0.05% (Gibco) for 10 min at 37 °C.
To obtain a single cell suspension, cells were pipetted through a 40-µm cell strainer 30 .
After centrifugation at 300g for 5min, cell pellets were resuspended with 10% FCS and fixed by adding an equal volume of 4% formaldehyde (Sigma-Aldrich) and mixed for 10 min at room temperature while shaking. Fixation was quenched using 1.425 M glycine (Merck) on ice and immediately centrifuged at 400g for 8 min. Pelleted cells were then resuspended in lysis buffer (50 mM Tris, pH 7.5; 150 mM NaCl; 5 mM EDTA; 0.5% NP-40; 1.15% Triton X-100; protease inhibitors (Roche)), and nuclei were pelleted again by centrifugation at 750g for 5min.
The pellet was washed with 1x DpnII buffer, resuspended in 50µl 0.5% SDS and incubated for 10min at 62°C. After that 145µl water and 25µl 10% Triton (Sigma) was added to quench the SDS. After a 37°C incubation, 25µl DpnII buffer and 100U DpnII was added. The digestion reaction was incubated for 2h at 37°C, after 1h another 10U were added. After the digestion, DpnII was inactivated at 65°C for 20min.
Universal sequencing adaptor were added using the NEBnext Ultra DNA Library Kit (NEB) according to the supplier's recommendation. Samples were sequenced with Ilumina Hi-Seq technology according to standard protocols and 75bp PE mode. 200 million reads were generated for IMR-5/75, 5 million reads per sample were generated for all other cell lines.
FASTQ files were processed using the Juicer pipeline v1.5.6, CPU version 34 , which was set up with BWA v0.7.17 35 to map short reads to reference genome hg19, from which haplotype sequences were removed and to which the sequence of Epstein-Stein-Barr Virus (NC_007605.1) was added. Replicates were processed individually. Mapped and filtered reads were merged afterwards. A threshold of MAPQ≥30 was applied for the generation of Hi-C maps with Juicer tools v1.7.5 34 . Knight-Ruiz normalization of Hi-C signal was used for Hi-C maps. Virtual 4C signal for the MYCN locus was generated by the mean Knight-Ruiznormalized Hi-C signal across three 5kb bins (chr2:16,075,000-16,085,000).
All samples were sequenced with the HiSeq 2500 (Illumina) technology according to standard protocols and with 8 million reads per sample.
Reads were pre-processed, filtered for artefacts and mapped to the reference genome GRCh37 using BWA-MEM as described earlier 26 . After removing the viewpoint fragment as well as 1.5 kb up-and downstream of the viewpoint the raw read counts were normalized per million mapped reads (RPM) and a window of 10 fragments was chosen to smooth the profile.

Whole-genome sequencing
Cells were harvested and DNA was extracted using the NucleoSpin Tissue kit (Macherey-Nagel GmbH & Co. KG, Düren, Germany). Libraries for whole genome sequencing were prepared with the NEBNext Ultra II FS DNA Library Prep Kit for Illumina (New England BioLabs, Inc., Ipswich, MA). Libraries were sequenced on a NovaSeq S1 flow cell (Illumina, Inc., San Diego, CA) with 2x150bp paired-end reads. Quality control, adapter trimming, alignment, duplicate removal as for ChIP-seq data. Copy number variation was called (Control-FREEC 36 11.4 with default parameters). Structural variants were called using SvABA 37  Assembly results were visualized with Bandage 0.8.1 (https://rrwick.github.io/Bandage) and Ribbon (no version available, https://github.com/MariaNattestad/Ribbon). CpG methylation was called from the unfiltered raw FAST5 files using Megalodon 0.1.0 (Oxford Nanopore Technologies Ltd., Oxford, UK).

Fluorescence in situ hybridization
Cells were grown to 200,000 per well in six-well plates and metaphase-arrested using Colcemid

Analysis of copy number data
Public data was downloaded. Samples that were described as MYCN-amplified in the metadata but did not show MYCN amplification in the copy number profile were excluded. In order to generate an aggregate copy number profile, the genome was binned in 10kb bins and number of samples with overlapping amplifications was counted per bin. Randomized copy number profiles were generated by randomly sampling one of the original copy number profiles on chromosome 2 and randomly shifting it such that MYCN is still fully included within an amplified segment. For class I-specific shuffling, e4 had to be included as well; for class IIspecific shuffling, e4 was never included on the randomly shifted amplicon. Empirical P-values for significant co-amplification were derived by creating 10,000 randomized datasets with each amplicon randomly shifted and comparing the observed co-amplification frequency to the distribution of co-amplification frequencies in the randomized data. Empirical P-values were always one-sided and adjusted for multiple comparisons using the Benjamini-Hochberg procedure.

Amplicon reconstruction
All unfiltered SvABA structural variant calls were filtered to exclude regions from the ENCODE blacklist 39 and small rearrangements of 1kb or less. As we were only aiming at the rearrangements common to all amplicons, we only considered breakpoints with more than 50 variant-support reads ('allele depth'). gGnome 40 was used to represent these data as a genome graph with nodes being breakpoint-free genomic intervals and edges being rearrangements ('alternate edge') or connections in the reference genomes ('reference edge'). We considered only nodes with high copy number, i.e. with a mean whole-genome sequencing coverage of at least 10-fold the median coverage of chromosome 2. Then, reference edges were removed if its corresponding alternate edge was among the 25% highest allele-depth edges. The resulting graph was then searched for the circular, MYCN-containing walk that included the highest number of nodes without using any node twice. We used gTrack (https://github.com/mskilab/gTrack) for visualization. For custom Hi-C maps of reconstructed amplicon sequences of CHP-212 and IMR-5-75, respectively, the corresponding regions from chromosome 2 were copied, ordered, oriented and compiled according to the results from the amplicon reconstruction and added to the reference genome. Additionally, these copied regions were masked with 'N' at the original locations on chromosome 2 to allow a proper mapping of reads to the amplicon sequence. The contribution of Hi-C di-tags from these regions on chromosome 2 to the amplicon Hi-C map is expected be minor, because the copy number of amplicons is much higher than the number of wild type alleles. Juicebox v1.11.08 was used to visualize Hi-C maps with a bin size of 5 kb and Knight-Ruiz normalization [41][42][43] .   c, Enrichment for co-amplification with MYCN of genomic regions on 2p (red, co-amplification more frequent than expected by chance; blue, co-amplification less frequent than expected by chance).