Sequence determinants of histone–DNA binding preferences: Comment on “Cracking the chromatin code: Precise rule of nucleosome positioning” by Edward N. Trifonov

Abstract

DNA in eukaryotic cells is packaged into compact chromatin state. The fundamental unit of chromatin is a nucleosome, a highly conserved protein-DNA complex in which ~ 147 basepairs (bp) of DNA are wrapped around a histone octamer in a left-handed superhelix [1]. Histone-DNA binding affinity depends on the nucleotide sequence of the nucleosomal site: indeed, the ability of the DNA molecule to bend into the superhelix is mostly governed by dinucleotide base-stacking energies. The review by Trifonov [2] emphasizes the role of deformational properties of DNA in nucleosome positioning and energetics, proposing a novel sequence motif CGRAAATTTYCG that favors nucleosome formation. The motif is inferred from a large-scale map of C.elegans nucleosomes [3]. In building up the case for the 10–11 bp-periodic nucleosome positioning motif shown above, the author has chosen to focus mostly on his own work and the work of his colleagues. However, I find myself intrigued by how well the proposed pattern stacks up against some of the other models and datasets (nucleosome positioning determinants and the idea of the “nucleo-some code” have recently garnered a lot of interest in the chromatin field). The author argues that due to steric exclusion between neighboring particles only single-nucleosome conformations can be used to compare experiment with theory [4]. However, techniques similar to dynamic programming in computer science and transfer matrices in physics can be used to convert histone-DNA binding energies into probabilities of nucleosome formation at every bp, without any approximations related to the finite particle size (see e.g. [5]). Furthermore, because the DNA bendability matrix proposed by the author is capable of placing nucleosomes with 1 bp resolution, only seven nucleosomes whose positions are known precisely from experiment have been chosen to test the model, with impressive success [4]. This seems to be too restrictive – certainly a high-resolution algorithm can predict lower-resolution data. Besides, the vast majority of the algorithms proposed in the literature also have 1 bp resolution and have nonetheless been used to predict genome-wide occupancy profiles for nucleosomes mapped with ~ 10–20 bp precision by micrococcal nuclease (MNase). Moreover, C.elegans data from which the model was inferred in the first place employed MNase digestion followed by 454 pyrosequencing [3], and therefore has the usual ~ 10 – 20 bp accuracy. It would be especially interesting to see whether the DNA bendability model proposed by Trifonov and colleagues has greater predictive power against large-scale nucleosome maps than simple models that assign the same scores to mono- and dinucleotides of the same type, regardless of their position with respect to the DNA helical repeat [6, 7]. Finally, nucleosomes on which the analysis by Trifonov and colleagues is based come from in vivo chromatin in a mixture of C.elegans cells. The nucleosome positions in this sample are averaged over cell types and may have been affected, among other things, by chromatin remodeling enzymes and competition with other DNA-binding proteins. It would be reassuring to see the proposed motif also appear in two recent large-scale maps of nucleosomes positioned in vitro on genomic sequences from S.cerevisiae and E.coli [8, 9].