Heli
Tan
ab,
Tuoqi
Liu
a,
Jiajun
Zhang
a and
Tianshou
Zhou
*a
aSchool of Mathematics, Sun Yat-Sen University, Guangzhou 510275, P. R. China. E-mail: mcszhtsh@mail.sysu.edu.cn
bSchool of Mathematics and Computational Science, Xiangtan University, XiangTan 411105, P. R. China
First published on 3rd November 2016
Chromosomal regions are often dynamically modified by histones, leading to the uncertainty of nucleosome positions. Experiments have provided evidence for this randomness, but it is unclear how it impacts epigenetic heritability. Here, by analyzing a mechanic model at the molecular level, which considers three representative types of nucleosomes (unmodified, methylated, and acetylated) and dynamic nucleosome modifications, we find that in contrast to the equidistance partition of nucleosomes, random partition can significantly enhance heritable bistability. Moreover, the more “chaotic” the nucleosome positions are, the better the heritable bistability is, in contrast to the previous view. In both cases of nucleosome positioning, heritable bistability occurs only when the total nucleosome number is beyond a threshold, and it depends strongly on the allocation rate that enzymes regulate transitions between different nucleosome types. Thus, we conclude that random positioning of nucleosomes is an unneglectable factor impacting heritable bistability. A point worth mentioning is that our model established on a master equation can easily be extended to include other more complex processes underlying dynamic nucleosome modifications.
Epigenetic mechanisms can be roughly divided into two classes: one class involves cis-specific and DNA-associated differences. For this class, the best understood example is alternative DNA methylation.11–14 The other class is determined by circuits of diffusible and trans-acting factors that can exist in alternative regulatory states through positive feedback.15–17 The best understood examples include stable alternative induction states of the lac operon18 and the phage lambda CI-Cro switch.19 In contrast to the former class, which is mostly clearly identified when the alternative states can coexist within the same cell,20 the latter class, which is more common in prokaryotic and eukaryotic cells and is therefore a major class of epigenetic mechanisms, has not been well identified. This paper will focus on the latter.
As a consequence of epigenetic mechanisms, heritable bistability may be complicated by various nucleosome modifications. It is known that nucleosomes can carry various chemical modifications (e.g., acetylation and methylation) at different amino acid positions on different histones, potentially endowing a large information capacity with each nucleosome. Specific additions and removals of these nucleosome modifications are carried out by several classes of enzymes such as histone acetyltransferases (HATs), histone methyltransferases (HMTs), histone deacetylases (HDACs), and histone demethylases (HDMs)21 (we will limit our discussion to the catalytic roles of these four enzymes in this paper). Importantly, positive feedback can arise due to the corresponding modification reactions, e.g., positive feedback is formed if nucleosomes carry a particular modification recruit enzymes, which in turn catalyze similar modification of neighboring nucleosomes.2 Enzyme recruitment leads to stochastic transitions between distinct types of nucleosomes. However, since some enzymes such as HATs, HDACs, and HMTs are known to associate in vitro or in vivo with histones of the type that they are capable of producing,22–25 this makes a cluster of nucleosomes still able to be kept in particular modification states. Annunziato suggested that these states are inherited through DNA replication,26–28 based on the fact that nucleosomes on the parental DNA strand are distributed to both daughter strands and the enzymes recruited by these parental nucleosomes may then establish the pattern of parental modification on the newly deposited nucleosomes.
Furthermore, the positions of nucleosomes are important factors affecting a number of vital biological processes.29–34 In fact, the eukaryotic genome is packaged into chromatin, which consists of a basic repeating unit of nucleosomes arranged in regularly spaced arrays. DNA sequence is not the sole determinant of nucleosome positioning since proteins often compete with nucleosomes for binding to specific DNA regulatory sites,35–37 and ATP-dependent chromatin remodelers actively displace nucleosomes.38–40 Small et al.41 developed an experimental method to examine nucleosome positions in single cells, and revealed significant heterogeneity of nucleosome configurations within a population and complexity of nucleosome positioning as well as its role in regulating gene expression. In spite of this, the exact positions of nucleosomes are not known. In this paper, we will consider random positioning of nucleosomes, focusing on its effects on epigenetic heritability.
Experimental evidence supports the fact that nucleosome positions can impact heritable bistability, but this impact was neglected in previous studies. In order to reveal the mechanism of epigenetic cell memory, Dodd et al.2 proposed a stochastic model, which assumed that in the recruitment reaction, any nucleosome can act on any other nucleosome in the region, and thus one kind of modification can “jump” over differently modified nucleosomes. Such jumping might be facilitated by a higher-order chromatin structure, by DNA-looping or by more complex processes. However, this model only considers that nucleosome positions are equidistance in a finite region. Until now, it seems to us that no one has investigated how random positioning of nucleosomes influences heritable bistability. In addition, some previously proposed models took into consideration the fact that contacts between nucleosomes decrease with increasing distance between them on the DNA, and utilized a power law to quantify this contact, i.e., assuming that the probability of contact is proportional to 1/d1.5, where d represents the distance between two neighboring nucleosomes.42,43 In this paper, we also adopt such a power-law contact model to quantify the impact of one nucleosome on another nucleosome adjacent to it.
In summary, heritable bistability is necessary for cells of identical genomes to maintain distinct functional identities important for multi-cell organisms, factors influencing this bistability are diverse, and the positions of nucleosomes are not fixed but random. To our knowledge, there has been no systematic analysis of how dynamic nucleosome modifications including nucleosomes' random positioning affect heritable bistability. In addition, although it is known from other systems that positive feedback (or double-negative feedback) is a necessary rather than a sufficient condition for bistability,16,44 it is not clear whether this is real for heritable bistability. In this paper, we will develop a mechanic model for epigenetic systems that have the potential to generate heritable bistability. This model, which considers not only nucleosome modification reactions but also nucleosome positioning (random or deterministic), can find its prototype in the silenced mating-type locus of the fission yeast Schizosaccharomyces pombe.2,45 Using this model, we examine how heritable bistability is generated, focusing on the effect of nucleosomes' random positioning. Through analysis, we found that random positioning of nucleosomes can strengthen heritable bistability in contrast to equal positioning, that there is a critical value of the total nucleosome number such that heritable bistability occurs only when the total nucleosome number is beyond this threshold, and that heritable bistability depends heavily on a rate between enzyme allocations. We emphasize that our mathematical model is established in terms of a chemical master equation, so it has many advantages, e.g., our model is easily extended to include more complex nucleosome modifications, thus being convenient for further studies.
Fig. 1 also shows the effect of nucleosome position on heritable bistability. We compare three representative cases of nucleosome positioning: the first case is that the distance between any two neighboring nucleosomes is equal (denoted by d the common distance), the second case is that the distance between two neighboring nucleosomes follows a uniform distribution with mean d, and the third case is that the distance between two neighboring nucleosomes follows a normal distribution also with mean d. From the left panel of Fig. 1, we observe that the bistable region for uniform distribution is larger than the one for normal distribution, which is larger than that for equal distance. This is because the line for uniform distribution is located below that for normal distribution whereas the latter is located below that for equal distance. This observation indicates that for a given allocation rate, the stronger the randomness of the distance between two neighboring nucleosomes is, the more easily heritable bistability is generated.
In addition, we also observe from the left diagram of Fig. 1 that the allocation rate between enzymes is an important factor affecting heritable bistability if the nucleosome number is fixed. In the next subsection, we will further investigate how this factor impacts heritable bistability from different angle points.
We also demonstrate time evolutions of the methylated nucleosome number in the right diagram of Fig. 1, where panels (B) and (C) correspond, respectively, to two points indicated in the left diagram of Fig. 1. These time evolutions correspond to the case where the nucleosome positions follow the normal distribution. This demonstration is mainly used to verify the correctness of the divided regions for bimodality and unimodality.
Throughout the following, we will fix the total number of nucleosomes at some value (e.g., N = 60) to characterize heritable bistability from two aspects: dwell time and epigenetic landscape. In each case, we will focus on the effects of the nucleosomes' random positions. Before that, we examine the effect of the allocation rate between enzymes on heritable bistability and compare the results obtained in three cases of nucleosome positioning.
A more important point is one we observe from Fig. 2, that different ways of nucleosome positioning leads to different effects of the allocation rate on heritable bistability. Specifically, three positioning ways cannot create heritable bistability for a small allocation rate (e.g., less than 3/7). With the gradual increase of the allocation rate (AR increases from 1/9 to 7/3), heritable bistability occurs first in the case of uniformly distributed distances, then in the case of normally distributed distances, and finally in the case of equal distances. In other words, the uniform distribution of nucleosome positions creates heritable bistability more easily than the normal distribution that creates heritable bistability more easily than the equal distribution. This implies that the random positions of nucleosomes promote heritable bistability in contrast to the deterministic positions, which is one main conclusion of this paper. In the next subsection, we will further verify this conclusion.
From Fig. 3(A), we first observe that the total times that methylated and acetylated nucleosomes dwell at m = 0 or a = 0 in three cases of distances are increasing functions of the allocation rate AR, and these functions are all convex, referring to Fig. 3(A). Second, the line corresponding to the case of uniform distribution is above the line corresponding to the case of normal distribution, which is above the line corresponding to the equidistance case. This implies that random nucleosome positions prolong the total resident time in contrast to equidistance. Furthermore, since the distances that follow uniform distribution are more uniformly distributed than those that follow normal distribution, the more “chaotic” the distances are, the longer is the time that the methylated or acetylated nucleosomes dwell at m = 0 or a = 0. These results are interesting on the one hand, and are in accordance with those obtained in the previous subsection on the other hand.
From Fig. 3(B), we observe that the line for the time evolution of the mean methylated nucleosome in the case of uniform distribution is above the line in the case of normal distribution, which is above the line in the equidistance case. This indicates that the more “chaotic” the distances between nucleosomes are, the larger is the mean number of methylated nucleosomes. Similarly, we observe from Fig. 3(C) that the line for the time evolution of the methylated nucleosome noise in the case of uniform distribution is above the line in the case of normal distribution, which is above the line in the equidistance case. This also indicates that the more “chaotic” the distances between nucleosomes are, the higher is the methylated nucleosome noise. By comparing Fig. 3(B) and (C), we find a counterintuitive phenomenon since the larger the mean is, the smaller the noise should be. In addition, we observe from Fig. 3(B) and (C) that there are sudden changes at the first time. Additionally, we point out that the results obtained by theoretical prediction [see eqn (10)] are in accordance with those obtained by numerical calculation (data are not shown) by noting b1,0 = 〈m〉 (see Section 3 for the definition of b1,0).
We define epigenetic landscape function U(m) as U(m) = −lnP(m), where P(m) represents the steady-state probability distribution of the methylated nucleosome number. We plot the dependence relationship of U(m) on both m and AR, and thus obtain Fig. 4, where three cases of distance distribution are shown: nucleosome distances of one kind follow uniform distribution, nucleosome distances of another kind follow normal distribution, and nucleosome distances of the other kind are equidistance. In Fig. 4, the deep blue area corresponds to low values of epigenetic landscape whereas the red area to high values. These three kinds of nucleosome distances can result in transitions from unimodal to bimodal peaks. For example, if the mean distance is set at d = 10 (in unit dp) (corresponding to the first row of Fig. 4), then the transition from the navy blue area to the light blue area (i.e., the transition from monostability to bistability) takes place at AR ≈ 0.5 for uniform distribution, at AR ≈ 0.9 for normal distribution, and at AR ≈ 1.2 for equidistance, indicating that the order for the easy occurrence of bistability is from the uniform distribution of distances to the normal distribution of distances and to equidistance. Similar phenomena exhibit in the case of acetylation, referring to the third row of Fig. 4 from which we observe that the shape of epigenetic landscape is basically similar to that in the case of methylation. These observations further verify from the view point of epigenetic landscape that random nucleosome positions result in heritable bistability more easily than equidistance on the one hand, and the more “chaotic” the nucleosome distances, the better (or the more robust) the bistability on the other hand.
We also investigate the effects of mean nucleosome distances on epigenetic landscape. Here, we show only a set of numerical results in the second row of Fig. 4, where the mean distance is set as d = 15. We observe that the transition from monostability to bistability takes place at AR ≈ 1.1 for uniform distribution, at AR ≈ 1.6 for normal distribution, and at AR ≈ 2.0 for equidistance. By comparing the diagrams in the first row and those in the second row of Fig. 4, we find that the mean nucleosome distances leading to bistability do not remarkably change the shape of the epigenetic landscape but different distance distributions result in different occurrences of bistability.
Second, we hypothesize that there are two different kinds of conversions: one is the so-called “recruited conversion”, namely, nucleosomes are actively interconverted by modifying and demodifying enzymes (HMTs, HDACs, HDMs, and HATs) that are recruited by the modified nucleosomes. It is this recruitment that forms the positive feedback necessary for the generation of heritable bistability. For clarity, we include four symmetrical positive feedback loops in the schematic shown in Fig. 5 and describe all active recruited processes in terms of one-rate parameter k+ that is absorbed in some functions fi. The other is the so-called “noisy conversion”, that is, nucleosomes are interconverted in a recruitment-independent manner. This “noise” in the system can be considered as due primarily to the activity of modifying enzymes that are either free or attached to nucleosomes beyond the region boundaries. In the absence of further information and for analysis simplicity, we also conclude four possible noise conversions in Fig. 5 with one-rate parameter k− that is also absorbed in some functions fi. We find that it is the ratio of the recruited and noise conversions that is critical in the system, so an independent noise parameter is not needed.
Third, we suppose that the nucleosome positions are random. In fact, no experimental evidence supports that they are fixed. In ref. 2 and 47 the authors theoretically studied how heritable bistability is generated in a simplified model, but they ignored the effects of nucleosome positions. Small et al.41 described an experimental technique to determine nucleosome positioning in single cells by virtue of the ability of the nucleosome to protect DNA from methylation, and showed that nucleosome positioning has assignable impacts on gene expression.
Finally, we assume that the rates of interconversion reactions at each nucleosome across the DNA region are the same (that is, homogeneous) with regard to nucleosome positions, although heterogeneity exists possibly in some systems, e.g., in the Schizosaccharomyces pombe mating-type system.45 However, our model can easily incorporate such heterogeneity.
Now, we quantify how one nucleosome influences another nucleosome. For this, we introduce a quantifier (denoted by ω), which is a function of distances between nucleosomes. For every nucleosome located in the middle of the region, we use ωi = 1/(di−1)p + 1/(di)p with i = 2, 3,…,N − 1 to quantify how its two neighbors impact this nucleosome, whereas for two nucleosomes located at the boundary, we use ω1 = 1/(d1)p and ωN = 1/(dN)p to quantify this kind of impact, where p is a positive parameter (since this parameter does not influence our qualitative conclusion, we will set p = 1.5 in our analysis. See ref. 2 for a reason of this setting). In the case that all di are deterministic, we assume that they are equal and denote by d the common distance. In the case that the distances are random, we assume that they are generated through some distribution with the mean d. In this paper, we consider only two common distributions: uniform and normal. Note that the larger the distances are, the smaller is the influence, so this setting of distances is in accordance with intuition.
Next, we consider the total impact, which is defined as
(1) |
(2a) |
(2b) |
(3) |
(4) |
When writing eqn (4) based on reactions, we do not consider the constraint on the total number of nucleosomes. If this total number is fixed, i.e., if m + u + a = N, then we can eliminate a variable from eqn (4). It should be noted that in some particular cases or under specific constraints (in fact, no experimental evidence supports that all the reactions listed in eqn (2) and (3) take place simultaneously. Rather, it is possible that only parts of them occur), eqn (4) can be reduced to simpler forms. Moreover, the reduced equations can have particular meanings, e.g., if only the reactions in eqn (2a) occur (this is equal to the case in which enzymes HDAC and HDM do not exist or equally we set f1 = 0 and f3 = 0), this corresponds to the modification process (the corresponding model is called the modification model); if only the reactions in eqn (2b) occur (this is equal to the case in which enzymes HMT and HAT do not exist or equally we set f2 = 0 and f4 = 0), this corresponds to the demodification process (the corresponding model is called the demodification model); if they occur simultaneously, the corresponding model is called the standard model.
Without loss of generality and by combining the above setting of nucleosome positions, we assume that transition rates for the four recruited conversions are given by f1 = f2 = mωi, and f3 = f4 = aωi, where ωi represents the distance impact factor (see Section 2.2) and can be generated by a distribution, i = 1,…,N. When one single ωi is used to quantify the distance influence on nucleosomes, we call the corresponding case a distance-based pathway for convenience. Thus, we have N such pathways in total. For the ith pathway, we denote by Pi(m,u,a;t) the corresponding joint probability distribution governed by eqn (4).
We introduce another impact factor, which is defined as the ratio of the one-rate parameter for recruited conversion over that for noisy conversion, that is, AR = k+/k−, to describe the noise in the system. For convenience, we call this factor the allocation rate, which can be used to convey the relative activities of the positive-feedback and noise-conversion processes. It should be noted that for a given AR, the actual ratio between the recruited and noisy conversions may vary, depending on the numbers of the three nucleosome types. Since there are no experimental data to estimate the value of AR, we let it change in a large range, e.g., 0 < AR < 100, so that the results to be obtained can include various biological possibilities.
Finally, the constraint m + u + a = N can eliminate one variable from eqn (4), resulting in
(5) |
(6) |
(7) |
Once all binomial moments bk(t) are found either analytically or numerically, we can reconstruct the distribution using these known bk(t). In fact, we have the following general reconstruction formula:
(8) |
(9) |
In addition, we can calculate the noise in nucleosomes according to the first two binomial moments, where the noise is defined as the ratio of the variance over the square of the mean. For example, for the methylated nucleosome M, the corresponding noise, denoted as ηM, can be expressed as
(10) |
Similarly, we can give formulae for calculating the noise in the number of unmodified nucleosomes.
Then, we calculate the potential landscape function Q(m,u) that is defined as Q(m,u) = −lnP(m,u) using P(m,u) obtained by the binomial moment method.
Third, we calculate U(m) = −lnP(m), where P(m) is the marginal distribution of P(m,u). Note that the landscape function U(m) should depend on the parameter AR (representing the allocation rate between enzymes that regulate transitions between different types of nucleosomes) since P(m,u) and further P(m) are all functions of AR.
The mechanism of how bistability is generated has been extensively investigated, especially in prokaryotes. Bistability in higher eukaryotes has also been obtained using, e.g., chimeric proteins derived from the viral protein VP16,49 but not using the endogenous genes under the physiological conditions considered here. The increasing knowledge about epigenetic mechanisms disagrees with a bistable model. In fact, extensive combinatorial possibilities of the dozens of histone modifications per nucleosome may produce very complex and fine-tuned outputs, very far away from the bistable model. Therefore, it is possible that random nucleosomal positioning enhances heritable epigenetic bistability, as shown in this paper. However, this random positioning can be deleterious for multistability situations which fit better with the behavior of endogenous genes in eukaryotes under physiological conditions and further investigations are needed.
In contrast to the existing related models, our proposed stochastic model for dynamic nucleosome modification has advantages in many aspects. We expect that this model can be extended relatively easily to include other nucleosome modification schemes, for example the 3-state model,50–52 and that our model can also be extended in a straightforward manner to include more than 3 types of nucleosomes, and the spatial positioning of nucleosomes,53–55 as well as heterochromatin resulting from the interaction between, e.g., histone H3/H4 N-termini and SIR3/SIR4 proteins.56
Our binomial moment approach is more powerful than the mean field approach57 in the sense that it allows one to explore probability distributions as well as epigenetic landscapes.58–61 The method can also deal with cases where the epigenetic system does not conform to a potential energy surface. For example, noise associated with cell divisions can be included in systems with double negative feedback between repressors, such as the CI-Cro feedback loop in the lysis-lysogeny switch of phage lambda.62
Our results are consistent with recent observations in mammalian cells in which increased cell division rates accelerated stochastic transitions between epigenetic states.63 This is because in our case, such acceleration means that the external impact (e.g., from random positioning of nucleosomes) is increased, leading to the enhancement of heritable bistability.
Finally, we argue two points: (1) our model considers the fact that the impact of nucleosome positions enters the model in a linear fashion, but nonlinearly entering manners are possible in real cases; (2) if more than 3 types of nucleosomes are considered, then multimodality may take place, depending on the cooperative role of nucleosome positions and the unique parameter—the allocation rate between enzymes regulating transitions between different types of nucleosomes.
This journal is © The Royal Society of Chemistry 2017 |