Reply to “Reassessing the abundance of H3K9me2 chromatin domains in embryonic stem cells”

Wen et al. reply:

Filion and van Steensel claim that the differentiation-related large organized chromatin K9 modifications (LOCKs) we reported¹ are not supported by our microarray data. We disagree, but also note that our conclusions regarding LOCKs were not based on array data alone but also on the many validations and functional experiments described in the paper, including real-time PCR validation, conservation of LOCKs, and genetic knockout of the histone methyltransferase G9a in ES cells and this knockout's influence on gene expression. Interestingly, these functional data are not challenged by Filion and van Steensel, nor do they question the existence of LOCKs or the tissue specificity of LOCKs. Figure 1a and Supplementary Figure 1 show replicate data, available on the web page cited in the paper (http://rafalab.jhsph.edu/k9LOCKs/), superimposed on the genome-wide data from the paper. The replicate data include two examples of undifferentiated ES cell lines cultured separately (biological replicates), as well as two differentiated cell types (differentiated ES cells and liver cells). In all cases where differences between undifferentiated and differentiated ES cells were described in the paper, the same differences are seen even more dramatically in the replicate experiments. Notably, the replicates are quite consistent for undifferentiated ES cells, and their signals were dwarfed by the relative signal in differentiated cells. The probability of observing by chance these same differences, with the same termini, is extremely low. We also performed quantitative real-time PCR validation of the LOCKs comparing chromatin immunoprecipitated DNA from undifferentiated and differentiated ES cells; this showed unequivocally that the LOCKs are differentiation specific (Fig. 1b).

Figure 1: Replicate data support differences of LOCKs between undifferentiated and differentiated ES cells.

(a) Comparison of genome-wide array and replicate array. Upper plot shows smoothed curves of undifferentiated ES (orange) and differentiated ES (green) and regions defined as LOCKs in the paper. Bottom plot shows curves of two independent ES cell cultures (blue and orange), differentiated ES cells (green) and liver (pink) in the same region. The custom array is as described in the paper. (b) qPCR validation of differentiated ES cell–specific LOCKs. The y axis shows ratios of ChIP/Input normalized to tubulin. Tubulin and MPos2 are negative and positive controls, respectively. MV1–MV3 are three regions in differentiated ES cell–specific LOCKs (MV1 and MV2 from LOCK at chr. 7: 7416826-9017990; MV3 from LOCK at chr. 1: 108835775-109160948). At least three replicate experiments were performed in each case. Primer sequences are available on request.

Regarding microarrays, Filion and van Steensel claim that variations between undifferentiated and differentiated ES cells are due to sample labeling or hybridization conditions. Our extensive experience with microarray data is that variation due to hybridization and labeling can be controlled by appropriately normalizing the data. Our group included expert statisticians and spent a great deal of time and thought on the statistical analysis. Our method was based on a completely data-driven procedure, as described in the Supplementary Methods section of the paper. Notice that the procedure and cutoff we used in the paper worked remarkably well at detecting locations with differential LOCKs between samples as confirmed by (i) RT-PCR for gene expression, (ii) association with gene expression in tissues and (iii) concordance with replicate experiments (ref. 1, Fig. 1 and Supplementary Fig. 1). Comparing various tissue-specific LOCKs using this same procedure revealed an extremely strong relationship between our statistical criteria of LOCKs and domain-specific gene silencing. We did consider other percentiles (cutoffs), and the conclusion that differentiated cells had more LOCKs did not change (Supplementary Fig. 2).

Note that the use of a two-state hidden Markov model (HMM) to assess specific microarray signals is not always appropriate. In our original study¹, we specifically said that the LOCKs detected in the genome-wide arrays are not necessarily absent in undifferentiated ES cells but may be minimally present compared to differentiated ES cells, as shown in our original Figure 3a. It is well known that ES cell cultures are usually contaminated with differentiated cells (typically 10% and often substantially more) even when clones are chosen for an apparently undifferentiated morphology^2,3,4—which, in fact, as we reported, we did not do. Kalmar et al. have now proven that ES cells are dynamically heterogeneous at the population level⁵. A more appropriate approach than fitting a two-state HMM is to fit an HMM with at least three states: baseline, LOCKs and apparent LOCKs due to underlying biology. It is clear that had Filion and van Steensel jointly fitted a three-state HMM, instead of a two-state HMM, to the undifferentiated and differentiated datasets, they would have obtained results very similar to ours.

Our experimental and statistical methods extend the boundaries of our ability to define differences in nuclear organization and are imperfect, just as is van Steensel's method for defining lamin-associated domains (LADs) through in vivo methylation by lamin fusion proteins⁶. But if our conclusions were wrong, why would the LOCKs we defined detect regions largely overlapping with LADs? van Steensel's group described changes in LADs during ES cell differentiation in an abstract at the Cold Spring Harbor Conference on Dynamic Organization of Nuclear Function (2008). There is even older evidence, although not mapped to specific chromosomal locations, showing an increase in H3K9me2 in differentiated cells compared to undifferentiated ES cells⁷. Furthermore, Bing Ren and colleagues have confirmed our observation of large heterochromatin domains of hundreds of kilobases in size arising in differentiated ES cells from regions with bumps of a few kilobases in undifferentiated ES cells, albeit in human cells and with different heterochromatin markers (ref. 8 and B. Ren, personal communication). They also showed that partially methylated domains (PMDs), in which DNA are less methylated in fibroblasts compared to human ES cells⁸, are enriched for expanded heterochromatin blocks in fibroblasts but not in ES cells. Interestingly, the LOCKs we defined in differentiated ES cells largely overlap the PMDs (Supplementary Fig. 3), even given that the mapping is cross-species.