In this work, we provide evidence that many translocation breakpoints observed in human diseases have high Hi-C contact frequencies in normal cells, suggesting a broad role for 3D chromatin organization in determining the frequency of translocations between partner loci. Previous data about this phenomenon is derived primarily from FISH 
, a technique that, although revealing, can investigate only a limited number of loci simultaneously. Chromosome conformation capture with Hi-C, in contrast, allows genome-wide, unbiased investigation of proximity-mediated interactions between translocation partners 
. Since Hi-C measures contact frequencies rather than average nuclear distances, this approach provides a test for the “contact first” model of genomic rearrangements.
We find that Hi-C detects frequent proximity of hundreds of translocation-prone loci, including 1) translocations recurrently observed by cytogenetics in primary cancers from multiple tissues, 2) unbiased collections of rearrangements detected in tumors by next-generation sequencing, and 3) translocations associated with rare Mendelian disorders. In all cases, we detect a subtle but significant enrichment for translocation-partner contacts when compared to a null distribution. Recurrent translocations from a matched cell lineage (hematopoietic malignancies) show a stronger Hi-C contact signal than translocations from tumors derived from other lineages, suggesting that tissue-specific chromosome conformation may contribute to rearrangement partner selection. We also identify several individual translocations that show particularly strong Hi-C contact frequencies, including t(12;19)(p13;p13) and t(4;14)(p16.3;q32.33), which are both recurrently found in multiple hematopoietic tumor types. We thus predict that these individual translocations, particularly those that do not involve the IGH locus (see below), have a particularly high probability of rearrangement due to their increased frequency of contact.
Although translocations often occur in areas of high Hi-C contact frequency, many translocations occur in areas of low or average Hi-C contact frequency. This may occur for several reasons. First, it is possible that even though the “contact first” model is responsible for the occurrence of some translocations, this signal cannot be detected in Hi-C data from a population of cells. Spatial genome organization may be unique to individual cells or sub-populations of cells, and thus aggregate Hi-C signal may fail to capture spatial interaction occurring in only a few cells (and it may be these cells that go on to form translocations under strong positive selection in the cancer).
Second, translocations that do not show an elevated Hi-C signal in our analysis may have occurred for reasons entirely unrelated to spatial co-localization. While we directly test the “contact first” model, spatial proximity is certainly not a sufficient condition for observing a translocation in disease genomes 
. Some translocations may occur under a “breakage-first” model, whereby cellular mechanisms exist to co-localize DSB ends following breakage 
. The frequency of observed translocation is also affected by multiple other factors, potentially including DSB susceptibility 
, DSB mobility 
, spatial heterogeneity among cells, positive selection, and ascertainment bias. To dissect the contributions of other cellular processes to translocation partner selection, investigators will need to examine concurrent genome-wide profiles of these phenomena in the same cellular system.
It is also worth nothing that while we detect a significant enrichment of spatial proximity between observed translocation partners, the datasets analyzed here were not sufficient to directly test whether the formation frequencies of these translocations are correlated with the contact frequencies between genomic partners, as has been shown experimentally for induced translocations 
. This is because the true frequency with which translocations form
is quite different from the frequency with which they have been observed
in clinical databases. Observed frequencies are biased by factors including positive selection during neoplastic progression, the limited sensitivity of current methods to detect rare rearrangements in clinical samples, and ascertainment biases intrinsic to non-whole genome testing and clinical sample collection. While we do, despite these factors, observe significant correlation between the number of reports of a translocation in the Mitelman database and the Hi-C interaction score (Figure S5
), we do not detect significant correlation (T
0.532) between the number of observations of a translocation and the Hi-C interaction permutation p
-value (the more robust measure of spatial proximity).
In lymphoid malignancies, the spectrum of observed translocations is drastically altered by the presence of the DSB-inducing enzyme AID, which contributes to the formation of rearrangements involving loci throughout the genome, particularly the Ig locus 
. The formation of DSBs in these regions may then be a dominant force in determining which loci rearrange. Our results suggest that proximity plays a role in the formation of translocations in other tumor types as well. Indeed, given the absence of AID in non-lymphoid tissues, proximity might play a relatively larger role in determining the landscape of observed rearrangements in non-lymphoid cancers. Thus we suggest that future investigations of spatial proximity in cancer will benefit from examination of chromatin architecture in non-lymphoid tissues.
In an important methodological demonstration, we show the utility of Hi-C data in the discovery and fine-mapping of existing translocations in malignant cells genome-wide (, Figure S1
), in accord with previous work using the targeted 4C method 
. This finding has two key implications for future genome-wide analyses of chromatin structure in cancer cells. First, Hi-C data is able to accurately detect translocation breakpoints, allowing genome-wide analysis of structural rearrangements. Indeed, Hi-C data may be able to reconstruct the complete karyotype of a cancer cell, including deletions, amplifications, inversions, and other chromosomal alterations. Second, this analysis will be critical in filtering out the effects of chromosomal translocations that might interfere with the study of other trans
proximity-mediated interactions in future Hi-C studies.
Given the role of spatial proximity in translocation partner selection demonstrated in this study, the molecular mechanisms that govern three-dimensional genomic architecture in normal and cancerous cells may prove important in our understanding of cancer etiology. Work to characterize the interactions between chromosome conformation and triggers for rearrangements will help to untangle the molecular processes of damage and aberrant repair that contribute to oncogenic transformation.