We observed that nucleosome positioning at promoter regions was similar between two unrelated strains of S. cerevisiae. Because these strains have a large extent of transcriptional differences, this argues that differences in nucleosome occupancy profiles are not a major source of intra-species variation in gene expression.
In contrast, the epigenomic profile of H3K14ac was highly variable and this variability targeted specific nucleosomes. The presence/absence of a modification at a particular nucleosome in a given cell is, by definition, a discrete state. However, we observed quantitative acetylation differences that were often subtle (1.2 to 1.5 fold). This is likely due to high cell-to-cell heterogeneity and high dynamics of the acetylated state: all states from billions of cells were averaged in our samples, and no dynamical information was acquired over time. It is therefore important to interpret SNEPs as differences in the overall acetylation level across a cell population and not as a uniform epigenotype shared by all cells of the sample.
Natural epigenetic variation was previously reported at the level of methylated DNA (meDNA), particularly in plants 
. In this case also, differences were not necessarily discrete but often continuous. Important properties of SNEPs distinguish them from meDNA epi-polymorphisms. Methylated epi-alleles were predictive of lower gene expression 
but SNEPs with reduced acetylation were not. In addition, no evidence was reported on a possible role of meDNA variation on the dynamics of gene activation.
Since histone-tail modifications are known to be highly reversible and dynamic, the basis and the origin of SNEPs remain to be further investigated. We observed that the two strains had different overall patterns of acetylation along genes, with a preferential acetylation near TSS and TES in the BY strain, while the RM strain had enriched acetylation in the second half of transcribed regions. This pattern difference accounted for many SNEPs and may result from trans-acting factors that act differentially in the two strains. However, 1806 SNEPs could not be attributed to this general inter-strain difference. Focusing on these SNEPs only, we looked again at their genomic distribution, their potential correlation to expression divergence and enrichment in genes with high plasticity (Figure S9
). All conclusions made in our study were retrieved for this subset of SNEPs. Thus, the differential pattern of acetylation does not explain the general SNEP properties. Nucleosomal epi-polymorphisms may offer an alternative to irreversible nucleotide mutations. How BYac SNEPs accumulated at regulatory regions of conserved DNA is unclear. As mentioned above, it may occurred with the fixation of a trans
-acting variation. Alternatively, accumulation may have occurred as a drift during laboratory culture conditions where fitness selection poorly applied. Future experiments examining a third wild strain will help determine if one of the two patterns is more ‘common’, if the stronger effect (acetylation fold-change) of BYac SNEP is peculiar, and if the abundance of SNEPs is similar in various pairwise comparisons of strains.
What is the origin of this epigenomic variability? E. Richards proposed a classification of epigenotypes based on their dependency on DNA variation 
, where the obligatory
qualifications relate to genetic controls that are full, absent or incomplete, respectively. Following this terminology, obligatory
SNEPs may result from genetic factors acting in cis
or in trans
. Known cis
-regulations are exemplified by position effects of transposable elements, rDNA repeats or telomeric sequences. Trans-acting genotypes may reside in histone acetyl-transferase or de-acetylase machineries, or in upstream regulatory factors. Such obligatory
SNEPs could have been fixed together with their genetic determinants. In contrast, if some SNEPs are pure
(independent of genotype) they likely result from their direct selection. As SNEPs seem to relate to the dynamics rather than the steady-state levels of gene expression, this selection may act through the ability to respond to environmental changes (the Baldwin effect). Also, interactions between epigenotypes and genotypes are expected since histone acetylation can modulate the buffering of cryptic genetic variations 
Acetylation of Lysine 14 of histone H3 at the beginning of protein-coding sequences has unambiguously been associated to high transcriptional activity in several studies 
. It is therefore surprising that a preferential acetylation in one strain is not accompanied by a higher gene expression. This illustrates the complexity by which the various layers of inter-strain molecular differences are connected. Previous studies showed that DNA polymorphisms act on transcripts abundance in a complex manner 
, with a large extent of gene x environment effects 
and that this genetic control was largely distinct from the control of protein levels 
. Our results show that chromatin histone-borne modifications provide yet another layer of diversity, with non-trivial connections to genotypes and transcripts levels. SNEP identification and characterization provide a basis for population epigenetics of histone-borne modifications, and future quantitative epigenetics studies such as previously suggested 
will define the nature of these dependencies, and their relevance to the control of complex traits.
The abundance of SNEPs in highly-responsive genes and our observation that one SNEP correlated with the dynamics of gene activation upon stimulation suggest a contribution to gene x environment interactions. This is in full agreement with a previous report describing the contribution of H3K27me3 at the FLC
locus of Arabidopsis
to natural variation in cold-induced acceleration of flowering 
. Except in such rare cases, gene-by-environment interactions have only been studied in the context of DNA variation. Integrating epigenotyping of histone marks in these investigations will likely better explain how individuals differ in their response to environmental changes.
In particular, attempts to predict and optimize the response to specific treatments is at the heart of personalized medicine. Chemical inhibitors of histone deacetylase are used in anti-cancer therapies and seem promising to fight other diseases 
, and ChIP-SEQ technologies 
will soon provide clinicians with epigenotyping possibilities. Our results suggest that histone modification profiles of human individuals may greatly differ, with likely consequences on treatment outcome.