Defining chromosomal translocation risks in cancer

Abstract

Chromosomal translocations are a hallmark of cancer. Unraveling the molecular mechanism of these rare genetic events requires a clear distinction between correlative and causative risk-determinants, where technical and analytical issues can be excluded. To meet this goal, we performed in-depth analyses of publicly available genome-wide datasets. In contrast to several recent reports, we demonstrate that chromosomal translocation risk is causally unrelated to promoter stalling (Spt5), transcriptional activity, or off-targeting activity of the activation-induced cytidine deaminase. Rather, an open chromatin configuration, which is not promoter-specific, explained the elevated translocation risk of promoter regions. Furthermore, the fact that gene size directly correlates with the translocation risk in mice and human cancers further demonstrated the general irrelevance of promoter-specific activities. Interestingly, a subset of translocations observed in cancer patients likely initiates from double-strand breaks induced by an access-independent process. Together, these unexpected and novel insights are fundamental in understanding the origin of chromosome translocations and, consequently, cancer.

Keywords: activation-induced cytidine deaminase; cancer; chromosomal translocation; multiomics; paired integrative analysis.

Fig. S1.

Translocations preferentially accumulate in transcriptional active genes. (A) Translocations as determined by HTGTS (1) are plotted against transcriptional activity. (B) Transcriptional activity of TL-genes in human lymphomas.

Fig. 1.

Promoter-specific activities do not rule translocation risk. (A) Genes in proximity to the primary break site (Myc) are highly predisposed to become involved in a translocation reaction. Translocation events were gradient-colored (red to black), where intrachromosomal (red) were distinguished from interchromosomal (black) translocations. The number of genes with a defined number of rearrangements were grouped and plotted against the number of rearrangements. The black-to-red gradient, therefore, represents the relative contribution of intrachromosomal rearrangements to each group, where red represents rearrangements on chromosome 15 where the I-SceI–Myc site locates. A small group of previously described genes (e.g., Pim1, IL4ra, and Cd83) revalidated as residing proximal to the DSB translocate (9), but most gene fusions arise distal. This approach enables a distinction between proximal- from distal-translocation events. The distal events, as defined in this analysis, are used in HTGTS-based analyses throughout this report. The asterisk signifies that when excluding intragenic rearrangements within the Myc locus itself (n = 66740), 74% of rearrangements arise within distal genes. (B) Large genes are predisposed to become involved in a chromosome translocation. The overall gene size distribution of genes (black) and translocated genes (red) are as found by the Mitelman database. (C) Gene size distribution of chromosome translocated genes as observed in HTGTS (mice) and cancer patients. The gene size distribution of distal HTGTS translocations is represented in the red curve (n = 4,532). To enable a direct comparison of the gene size distribution of translocated genes in humans and mice (HTGTS), all human genes were binned according their gene size, after which the fraction of 1,000 translocated genes as found in human cancer patients per bin was determined (black curve). The same was done for all mouse genes and 1,000 randomly selected translocated genes as found in HTGTS (orange curve). Apparently, human and mouse chromosome translocations follow a same size distribution and, therefore, similar selection criteria. (D) Gene size directly determines the chromosome translocation risk. If promoter-specific activities dominate the translocation risk, the number of rearrangements should not relate to gene size (orange line, expected slope = 0). In contrast, if promoter-specific activities basically do not contribute, doubling the median size of translocated genes is expected to double the translocation risk (expected slope = 1). If both parameters contribute, an intermediate slope is expected (slope >0 and <1). Remarkably, and only confirming the complete irrelevance of promoter-specific and -proximal activities, a linear increase was observed where the number of rearrangements exactly doubled when doubling the gene size (black line). Distal translocated genes were binned according to rearrangement frequency and the median length was determined for each bin for both the distal Myc target genes, as obtained with HTGTS (black, n = 4,353, distal targets in C), and genes randomly selected based on their length (red, n = 4,362). The rearrangement frequency increases linearly with size. Doubling the gene size doubles the rearrangement frequency (δ 1, 2, 3). Of note, the y axis intercept does not represent a fixed size, but rather relates to the number of translocations that accumulated in this specific HTGTS study. The red line represents the expected curve if translocated genes would be selected only on the basis of their gene length (i.e., nucleotide content). (E) Gene size distribution of distal genes that rearrange with Myc. Gene size distribution of genes that rearranged one, two, three, and four times during HTGTS. Note, a peak at gene size 9 (indicative for a promoter preference) is lacking in all distributions.

Fig. S2.

Gene sizes vs. translocations. (A and B) Promoters are in contrast to gene bodies of a more confined size. (C) When a DSB, or two DSBs, in large genes rearrange intragenically, no gene fusion will result, whereas the opposite holds for small genes (D).

Fig. 2.

Transcription does not facilitate chromosomal translocations in HTGTS. (A) Each translocated promoter was paired with a randomly selected promoter with equal mean Dam value. Hereafter, the difference in the GRO score was determined and subsequently binned for two independent replicates (red, TL). The same procedure was followed for randomly selected genes (gray, R). Only if translocations preferentially occur in transcribed promoters, transcriptional activity of a randomly selected promoter selected from the TL group is expected to be higher compared with a promoter selected from the NTL group (after matching both promoters on access). However, this is not the case (red). As an additional control, we also access-matched randomly picked promoters from the complete set of promoters (i.e., TLs and NTLs), with translocated promoters. This did not show a significantly (Wilcoxon test) different result from the previous analysis. (B) As in A, but now each translocated promoter was paired with a randomly selected promoter with equal DHS. Hereafter, the difference in the GRO score was determined and subsequently binned for six independent replicates (red, TL). The same procedure was followed for randomly selected genes (gray, R). (C) Each translocated gene was paired with a randomly selected gene with an equal gene size and Dam score. Subsequently, the difference in the FPKM-value of each pair was determined and binned based on their Dam score (red). The same procedure was followed for random genes (black). (D) Each translocated gene was paired with a randomly selected gene with equal access as measured by DHS profiling. Subsequently, the difference in the transcriptional activity score (in FPKM) was determined (red, TL). The same procedure was followed for randomly selected genes (gray, R).

Fig. S3.

Preferential binding of AID at stalled promoters has not been measured. (A) Schematic representation of Dam methylation profiling as a measure for access. (B) Correlation between two independent Dam experiments. (C) Relative transcriptional activity as measured by GRO-Seq, at CpG-poor (LCP, low-CpG content) promoters and CpG-rich (HCP, high-CpG content) promoters. (D) Red, CpG methylation levels as measured by MethylCap; blue, GATC Dam methylation levels as measured by Dam profiling. Window range: −40 kb and +40 kb from the transcription start site (TSS). (E) Red, CpG methylation levels as measured by MethylCap; blue, GATC Dam methylation levels, as measured by Dam profiling. Window range: −2 kb and +2 kb from the TSS. (F) Based on the authors’ definition of AID targets and promoter classification (5), it appeared that AID would have a preference for stalled promoters. However, when excluding unclassified promoters from the analysis (G), only a small preference remained, which in the absence of true biological replicates lacks any significance (5). (H) In contrast to ChIP seq, DamID makes use of an internal reference Dam (50). AID activity as measured by a mutant GFP reversion assay (51) for Dam-only, (I) AID-Dam-AID, and (J) AID-Dam. After overexpression, AID in AID-Dam and Dam-AID fusions remained enzymatically active. (K) M-A plot of AID-Dam signal over Dam-only signal did not reveal AID specific signal. Red dots represent probes covering the IgM locus, the prime AID target.

Fig. 3.

Gene access facilitates chromosomal translocation in HTGTS. (A) Integrative analyses of DHS profiles revealed translocated genes as easily accessible. Each translocated gene was paired with a randomly selected gene with equal size and activity, as measured by RNA-Seq. Hereafter, the difference in access, as measured by DHS, was determined and binned for six independent replicates (red, TL). The same procedure was followed for randomly selected genes (gray, R). (B) Integrative analyses of Dam profiles reveal translocated genes as hyperaccessible. Active genes were selected [log₂(FPKM + 1) > 15] and hereafter each translocated gene was paired with a randomly selected gene, with an equal gene size and transcriptional activity. Then, the difference in the Dam value of each pair was determined and binned based on their Dam score (red). The same procedure was followed for random genes (black). (C) In addition, inactive translocation partners are more accessible compared with their controls. The median Dam score in promoters was determined for the following four groups: nonactive and NTL, nonactive and TL, highly active and NTL, and highly active and TL. (D) The promoters of translocated genes are highly accessible. Each translocated promoter was paired with a randomly selected promoter with equal activity as measured by GRO-Seq. Then the difference in the DHS score was determined and subsequently binned for six independent replicates (red, TL). The same procedure was followed for randomly selected genes (gray, R). (E) As in D, but now access was determined by Dam profiling. (F) Translocation partners as identified by HTGTS are readily accessible. Nine Dam-score groups were defined and the median gene size in which these reside was determined for each group (red, 1–9) and compared with the median gene size of translocation partners as identified by HTGTS (TL, blue), as well as randomly selected genes (R, black). Accessible elements appear to accumulate in genes with a median gene size of 10.9 (∼54,000 bp).

Fig. S4.

Redefining AID targets. (A and B) ChIP-Seq analysis of AID occupancy at active and inactive promoters. The promoter alignments of AID ChIP-Seq reads from activated B cells of (A) Aicda^−/− (a single IP) and (B) Aicda^+/+ mice (a combination of three IPs) is shown (12). Blue, CpG-rich HCP promoters are generically active in differentiated cells. Red, CpG-poor LCP promoters are generically silent in differentiated cells. (C) Promoter alignments of the methylation footprint of Dam and AID-Dam. For Dam-profiling we followed the DamID protocol as described in detail elsewhere (50). The protocol was applied on the Ramos Burkitts lymphoma B-cell line. The fragments were hybridized on Human ChIP 2.1 M Economy whole-genome Nimblegen tiling arrays (Roche) (D). (E) The original HTGTS Aicda^+/+ library (gray) was normalized (blue) to resemble the library composition of the Aicda^−/− library (red). (F) Balanced distribution of chromosome rearrangements in Aicda^−/− and Aicda^+/+settings. After normalization, the library composition, and multiple testing only a handful of “AID-dependent” targets remained. Green dots represent significant differential rearrangements between Aicda^−/− and Aicda^+/+ (P < 0.05). Blue dots indicate translocated genes that survived multiple testing (q < 0.05). Red circles are significant genes located on chromosome 15, on which Myc resides. Of note, three previously proposed AID targets—Pim1, Il4ra, and Ly6e—appeared differential for the Aicda^+/+ condition. Because these genes localize proximal to IgH, AID off-targeting is likely restricted to Ig regions, where AID is normally active (9). Extensive genome-wide AID off-targeting and activity appears not existing or irrelevant in determining the frequency and number of genome-wide chromosome translocations.

Fig. S5.

Differences in gene size of access-dependent and -independent translocations. Closed chromatin (A) and open chromatin structures (B) are equally prone to DSB formation processes involving small DNA damaging agents and irradiation. This is in contrast to large DNA modifying complexes targeting closed (C) and open (D) chromatin structures. (C and D) Chromatin access restricts to DSB formation and rearrangement processes involving large DNA damaging agents. (E) The median gene size of randomly selected genes will reflect the overall median gene size distribution of 9.1 (9,000 bp). The median gene size of genes, randomly selected based on their length, results in a median gene size larger than the overall median gene size of 12.0 (∼165,000 bp).

Fig. 4.

Translocation events can arise access-independent in cancers. (A) The shortest distance of each intrachromosomal translocation was determined, grouped according to the indicated distances windows (5–8, 8–11, 11–14, 14–17, 17–20), and plotted against the sum of the sizes of both translocation partners. (B, panel I, Left) Size distributions when genes were randomly selected if gene size is irrelevant [i.e., occurrence (O)]. (Panel I, Right] First, 100 genes were randomly selected based on their nucleotide content. Next, the nearest by gene, located on the same chromosome, was selected. The size distribution of the most adjacent neighboring genes was plotted. (Panel II) Translocating genes were binned based on their log gene sizes into five categories: S1, 5–7; S2, 7–9; S3, 9–11; S4, 11–13; S5, 13–15. The size distribution of their partners was plotted. (Panel III, Left) Random selection of genes based on their nucleotide content (R). (Panel III, Right) Size distribution of the translocation partners of RUNX1 (R1).

Fig. S6.

Consequences of unequal library preparation on sequence library compositions. (A) When performing an IP on cross-linked DNA, accessible genomic elements (shown in red) like promoters, accumulate background signal, whereas inaccessible regions remain underrepresented (49). (B) When combining three independent IPs (12), background signal will preferentially accumulate in accessible regions, while leaving the signal in inaccessible genomic elements relatively low. Notably, a linear normalization did not reveal any AID binding site, as shown previously (10).

Fig. S7.

AID translocation models. (A) Two-break model of how AID would contribute to AID-dependent chromosomal translocations if AID off-targeting exists and also induces a DSB at the off-target. (B) Proposed one-break model of how AID would contribute to AID-dependent chromosomal translocations if AID off-targeting does not exist (A-EG, alternative end generation).