Response to Casellas et al

Casellas and colleagues challenge the conclusions from our recent paper and claim that AID-dependent translocations occur independent of Igh proximity. In contrast to our study (Rocha et al., 2012), they focus solely on hotspots, which they claim represent the only ‘true’ AID mediated translocations (Hakim et al., 2012). However, their contention that other sites with translocation capture (TC) reads reflect AID independent rearrangements is not supported by any genome wide subtraction analysis examining signal enrichment in the Igh^I-SceI AID sufficient versus deficient sample. Nonetheless, they state that this is precisely the analysis that is missing from our study and that without this we could not distinguish between AID dependent and AID independent events. By implication, as per their suggestion we should find considerable overlap and comparable signal strength in the two data sets, which would account for the 90% of sites that they claim are AID independent translocations. Furthermore, non-overlapping regions or sites with signal enrichment in the AID sufficient data set should be restricted to AID induced hotspots. This is not what we observed.

First, one issue confounding a genome wide analysis of AID dependent translocations is the huge variation in the number of translocations identified in the two Igh^I-SceI AID deficient biological replicate data sets: 67,156 translocations for replicate # 1 and 1,247 translocations for replicate # 2 (Table S7 (Klein et al., 2011). Using 100kb windows we found a very low correlation between these controls (r=0.0768), indicating that any analysis of AID dependent translocation events could vary tremendously depending on which data set was used (Figure S1A). For this this we considered sites with >2 reads and in order not to skew our analysis we excluded the 350kb region around the I-SceI site where the highest frequency of translocations occur in both AID sufficient and deficient samples (Klein et al., 2011). Second, reads in the Igh^I-SceI AID deficient replicate # 1 are distributed fairly evenly across the genome with a signal enrichment that is uniformly low, in contrast to the AID sufficient sample which shows more localized enrichment of translocations (Figure S1A). Indeed, using the same exclusion criteria as outlined above, even in the absence of translocation hotspots we found a mean signal enrichment of 6.49 reads versus 11.97 in the AID deficient versus sufficient sample (t test: p <1e-05). Moreover, if we normalized for the total number of reads in the two samples these differences would be even greater because less weight would be attached to the events in the AID deficient sample, which contains the most reads. Third, it is not possible to perform a subtraction analysis between AID sufficient and deficient samples because the events in these two data sets are predominantly non-overlapping: 732 regions with signal content in the Igh^I-SceI AID sufficient sample were not represented at all in the AID deficient sample, while 15,127 regions with signal content in the Igh^I-SceI AID deficient sample were not represented at all in the AID sufficient sample. In Figure S1B we show a scatterplot comparing the signal in 100Kb windows in the Igh^I-SceI AID sufficient versus deficient data set to demonstrate the low correlation between the two samples (r=0.0405). In contrast, there is a high correlation between AID sufficient replicates (r=0.6492) demonstrating the validity of this approach. Additional assessment using other measures of associations (ROC analysis and a Fisher exact test) could not reject the null hypothesis that the translocations in the deficient and sufficient mice are statistically independent. Taken together these data indicate that, outside of hotspots and the region around the bait, the majority of translocations in the AID deficient data set are not represented as translocations in the AID sufficient data set.

Concerning other issues raised in the letter:

It is clear that 200 kb windows centered on TSSs is the preferred way to analyze interactions with specific genes rather than non-overlapping windows, as the former provides a unique signal for each gene examined. Nonetheless, we did not claim that this alternative method led to global differences in our results but suggested this may point to localized variation. However, we stressed the importance of different analytical methods for specific questions and showed that compared with the fixed window size, a domainogram approach is a more suitable representation of proximity as judged by the gold standard, DNA FISH.

We did not use 20kb windows to analyze interactions across the genome, as implied by Casellas et al., because we are aware that many genes would have no HindIII sites in this size window. Indeed, our 20kb analysis was limited to a region containing three hotspots with sufficient 4C resolution.

We treated the bait chromosome separately to the rest of the genome because interactions with the cis chromosome occur at higher frequency than interactions in trans and must be analyzed separately as recognized by pioneers of the 4C technique (van de Werken et al., 2012). In their letter Casellas et al., contend that we used an arbitrary cut off of 60 Mb from Igh for our analysis of the bait chromosome. This was not the case. We simply demonstrated that within ‘a linear distance of 60Mb, the strength of interaction decreases with increasing linear separation’.

In sum, using distinct complementary approaches 4C-seq/TC-seq and DNA interphase/metaphase FISH analyses we conclude that close proximity to Igh is a contributing factor to AID mediated translocations. However, we very clearly stated that ‘our data do not imply that every gene associated with Igh will be hit by AID as we know that there are many gene intrinsic factors that influence AID targeting’. Furthermore, we do not undermine the contribution of break frequency in determining translocation outcome, and say that, it is not surprising that Hakim et al., do not find a high degree of correlation between Igh proximity and translocation frequency in the hotspot subset because the frequency of double strand breaks will always be the rate limiting factor. Nonetheless, it is worth noting that out of the 234 AID mediated translocation hotspots associated with the Igh^I-SceI site identified by Klein et al., 95 (40%) were detected in cis on chromosome 12, while only 26 (11%) were identified on chromosome 15. In contrast, for AID mediated translocations associated with the Myc^I-SceI site on chromosome 15, the number of translocations on chromosome 12 dropped to 10%, while translocations on chromosome 15 increased to 33% (Table S4 (Klein et al., 2011). Since the location of AID generated breaks is not dependent on the location of the I-SceI site, these changes can only be explained by differences in the frequency of interactions in cis versus trans (Lieberman-Aiden et al., 2009; Lin et al., 2012), which give rise to a higher number of translocation hotspots in whichever chromosome contains the I-SceI site. These data provide clear support for the notion that proximity is a key determinant of AID-dependent translocation hotspots.