BAYESIAN NETWORK ANALYSIS OF RELATIONSHIPS BETWEEN

NUCLEOSOME DYNAMICS AND TRANSCRIPTIONAL

REGULATORY FACTORS

Bich Hai Ho, Ngoc Tu Le and Tu Bao Ho

Japan Advanced Institute of Science and Technology, 1-1 Asahidai, Nomi, 923-1292 Ishikawa, Japan

Keywords:

Nucleosome dynamics, Bayesian network, Post-translational histone modiﬁcation, Transcriptional regulator.

Abstract:

Intergenic regions are unstable, owing to trans-regulatory factors that regulate chromatin structure. Nucleo-

some organization at promoter has been shown to exhibit distinct patterns corresponding to the level of gene

expression. Post-translational modiﬁcations (PTMs) of histone proteins and transcriptional regulators, includ-

ing chromatin remodeling complexes (CRCs), general transcription factors (GTFs), and RNA polymerase II

(PolII), are presumably related to the establishment of such nucleosome dynamics. However, their concrete

relationships, especially in gene regulation, remain elusive. We, therefore, sought to understand the functional

linkages among these factors and nucleosome dynamics by deriving a Bayesian network (BN)-based model

representing their interactions. Based on the recovered network, learnt from 8 PTMs and 15 transcriptional

regulators at 4034 S.cerevisiae promoters, we speculate that nucleosome organization at promoter is intention-

ally volatile in various regulatory pathways. Notably, interactions of CRCs/GTFs and H3 histone methylation

were inferred to co-function with nucleosome dynamics in gene repression and pre-initiation complex (PIC)

formation. Our results afﬁrm the hypothesis that extrinsic factors take part in regulating nucleosome dynamics.

More thorough investigation can be made by adding more factors and using our proposed method.

1 INTRODUCTION

Eukaryotic genomes are packaged inside cell nucleus

under chromatin structure like a bead-on-string ﬁber

of nucleosomes. As a fundamental unit, nucleosome

contains a core of octamer histone proteins wrapped

around by 147bp of DNA (Luger et al., 1997). More

than DNA packaging, chromatin involves in various

cellular processes such as transcription, DNA replica-

tion, etc., by occluding the access of biological ma-

chineries to cis-regulatory elements and/or modify-

ing related epigenetic information. To overcome the

obstacle imposed by chromatin, cells have developed

complicated pathways (Li et al., 2007), in which nu-

cleosome must be dislocated from chromatin to pro-

vide access to the underlying DNA sequences. While

positioning are strongly inﬂuenced by intrinsic DNA

sequence preference (Kaplan et al., 2010), the rear-

rangement can be ﬂexibly modulated by extrinsic fac-

tors, e.g., DNA-binding factors and CRCs (Wan et al.,

2009). They help to maintain the periodicity, hence

the corresponding transcription activities, by directly

altering nucleosome organization in various manners

(Venters and Pugh, 2009). PTMs were shown in vari-

ous works to be related to nucleosome spatial organi-

zation (Cui et al., 2010). These two factors interact in

that PTMs serve as their targeting marks, and in turn,

the locations of PTMs is modulated by those regula-

tory proteins. Such series of highly regulated interac-

tions may necessarily be characterized by a network

featuring variable correlations. Since the data in use

here are all related to transcriptional activities, we ref-

erenced to the common resulting effects on gene ex-

pression to infer possible functional linkages.

Taken together, we speculated that in complex in-

teraction network, the nucleosome dynamics at pro-

moter may play an intermediate role, i.e. affecting as

well as being affected by other factors, in regulatory

pathways, assuming that there are two classes of pro-

moters, unstable with periodic nucleosome arrange-

ments, and stable without. We employed Bayesian

network (BN), a class of probabilistic graphical mod-

els that can capture not only co-occurrence pattern

but also interaction/dependency among variables, for

interaction modeling. BN has been used to recon-

struct many kinds of cellular networks, such as gene

regulatory networks and protein interaction networks

(Friedman et al., 2000). Compared with previous

299

Ho B., Le N. and Ho T..

BAYESIAN NETWORK ANALYSIS OF RELATIONSHIPS BETWEEN NUCLEOSOME DYNAMICS AND TRANSCRIPTIONAL REGULATORY

FACTORS.

DOI: 10.5220/0003773502990302

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (BIOINFORMATICS-2012), pages 299-302

ISBN: 978-989-8425-90-4

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

ﬁndings, we showed that the built network can re-

cover known relationships as well as functional link-

ages consistent with knowledge about various DNA-

mediated processes.

2 METHOD

2.1 Datasets

The experimental data of 24 features Saccharomyces

cerevisiae were gathered as follows: Two classes

of 4043 yeast promoters, including 1355 unsta-

ble/dynamic (no periodicity and high expression) and

2688 stable (explicit periodicity and low expression)

were obtained from (Wan et al., 2009); 8 PTMs (from

−1000bp to TSSs) were obtained from (Pokholok et

al., 2005); 15 transcriptional regulators including 6

GTFs, 1 PolII component, and 8 CRCs from (Ven-

ters and Pugh, 2009). Data was discretized using

proportional k-interval discretization (PIKD) (Yu et

al., 2008) with intervals of [33%, 66%], [20%, 80%],

[40%, 60%], and determined by K-means clustering

(k = 3). As a result, [20%,80%] proportional 3-

interval scheme gave us the most reasonable network

(data not shown).

2.2 Bayesian Networks

2.2.1 Deﬁnition

A Bayesian network for a set of variables X =

,... , X

} is a probabilistic model consisting of

two components (Heckerman, et al., 1995):

• A network structure S, which is a directed acyclic

graph, representing conditional (in)dependence

relationships among variables in X

• A set P of local probability distributions associ-

ated with each variable.

Because the main target of our work is to uncover

the relationships among the PTMs so we are inter-

ested in the problem of learning BN structure. We

employed the score-based search method (Jensen and

Nielsen, 2007) to learn a BN structure representing

relationships among PTMs. To score a candidate net-

work, we used a Bayesian scoring metric, namely

BDe (Bayesian metric with Dirichlet prior and equiv-

alence) (Heckerman, et al., 1995).

2.2.2 Bootstrapping and Selection of Cut-off

Threshold

The search-and-score method generates a different

network on each run, and only one with highest score

is output. Hence, we employed the bootstrapping

method (Friedman et al., 2000) to estimate the con-

ﬁdence level for each edge in the resulting network.

Then following hybrid method was proposed to de-

rive a suitable threshold value for conﬁdence level of

each edge in the resulting network:

1. Divide data into n datasets using n-fold cross-

validation, t times.

2. At each time, n bootstrapped consensus BNs

.. . N

are output, using a ﬁxed τ. Then, n cross-

validated networks are combined into one by in-

cluding edges agreed by θ graphs. Thus, with

each pair of (τ and θ), one ﬁnal network N is

learnt.

3. To measure the goodness of the learnt network,

accuracy (acc) and coverage (cov) are used and

plotted as receiver operating characteristic (ROC)

curve. Co-ordinates of each point in the curve is

the average of t times (with standard deviation).

The chosen network has τ and θ that generate the

largest area under the curve (AUC).

acc

♯(N

∩ N)

♯N

,cov

♯(N

∩ N)

♯N

,i = 1, .. ., n

(1)

where the numerator is the number of overlapping

edges of each network with the common one; the

denominator is the number of edges in each net-

work and the common one, respectively.

In our experiment, we split data into 5 datasets (ac-

cording to 5-fold cross-validation) for t = 50 times;

on each, we bootstrapped 100 times (m= 100), which

totally resulted in 25, 000 input datasets for learning.

Each edge in the consensus network has a related con-

ﬁdence score, measured by the number of times it ap-

pears in 100 bootstrapped ones. Threshold τ was cho-

sen in the range of [0.5, 0.8] with the step of 0.05. The

combined network consisted of overlapping edges by

θ = {2, 3,4} cross-validated ones. We then chose pa-

rameters θ = 3 and τ = 0.65 to produce a network

with 24 features as its nodes and 36 edges represent-

ing the functional linkages among PTMs, transcrip-

tional regulators, and nucleosome dynamics (Fig. 1).

3 RESULTS AND DISCUSSION

3.1 Network Recovers Reliable

Functional Linkages

Comparing our network (Fig. 1) with previous ﬁnd-

ings, we found a remarkable consistency with original

results reported in (Venters and Pugh, 2009). CRCs

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

300

Figure 1: PTMs, TFs, and nucleosome dynamics network with conﬁdence level (in range [0,1]). Note that the edge direction-

ality does not show deﬁnite causality, due to the nature of the learning algorithm; thus in most cases, we inferred correlations

or interactions among network nodes.

with subunits such as Swr1, Ioc2, Ioc3, Ino80, etc.,

appear in the upper part of the network, showing

causal relationships towards GTFs, which generally

reﬂects the role of CRCs as an ATP-dependent nucle-

osome translocator giving way to GTFs accessing the

buried DNA. From the network topology, it is clear

that nucleosomedynamics bridges two groups of tran-

scriptional regulators and PTMs, for which, unfortu-

nately, we have yet to ﬁnd direct evidence. Nodes

of high out degree (number of edges pointing away

from node), such as Swr1 (out degree of 6) and Ssl1

(out degree of 5), may play a central role; Swr1, (sub-

unit of SWR-C), is implicated in the deposition of

histone variant H2A.Z in promoter, which provides

a molecular mechanism to regulate transcription and

DNA repair (Luksend et al., 2010). Especially, to-

gether with NuA4, an essential histone acetyltrans-

ferase, they function in NuA4/SWR-C/H2A.Z pathway

to regulate chromosome stability.

3.2 Nucleosome Dynamics in

Regulatory Processes

3.2.1 PIC Formation by GTFs and PolII Affects

Nucleosome Dynamics

It is agreed that GTFs assemble at promoter to form

pre-initiation complex (PIC). The interactions among

them have been well investigated, and the related sub-

graph (TFIIH(Ssl1) → TFIIE(Tfa1) → THIIE(Tfg1))

→ TFIIB(Sua7) → TBP) → PolII (Rpo21) → State

in our network is consistent with previous ﬁndings

(Samorodnitsky and Pugh, 2010). Especially, these

factors were considered in (Samorodnitsky and Pugh,

2010), a modeling work (PathCom) towards the or-

dering of PIC assembly/disassembly at the genes of

yeast. The proposed assembly order was recruit-

ment of TBP, TFIIB ﬁrst, then PolII and other GTFs

(TFIIB, TFIIE, TFIIF, TFIIH in order); disassembly

goes backwards. Interpretation from our network that

the binding of factor at child node correlates with that

of parent node in an order manner, i.e., the biding

of parent affects that of child, not vice versa, is rea-

sonably consistent with this model. Hence, from our

network, nucleosome organization (node State) may

BAYESIAN NETWORK ANALYSIS OF RELATIONSHIPS BETWEEN NUCLEOSOME DYNAMICS AND

TRANSCRIPTIONAL REGULATORY FACTORS

301

inferably alter along with PIC forming (most impor-

tantly, PolII), which well matches with the known hy-

pothesis that nucleosome dynamics facilitates access

to GTFs (Morse, 2007) and CRCs.

3.2.2 Limited Gene Expression by CRCs is

Related to Periodic Nucleosome

Organization at Promoter

We analyzed the chain of SWR-C(Swr1) →

INO80(Ino80) → ISWIa/b(Isw1) → PolII(Rpo021)

→ State to understand the mechanism of gene repres-

sion, which is observable in our data as the periodic

nucleosome organization of limitedly expressed

genes. As pointed out in (Lindstrom et al., 2006),

Isw1 shows parallel functions with NuA4 and Swr1

complexes in repressing genes. ISWa/b plays a role

in repressing transcription (Pinskaya et al., 2009).

Hence, the interactions of Swr1, NuA4, Isw1, and

Rpo21 at promoter can reasonably reason the gene

repressing directionality. In our network, however,

INO80 stands as a bridge between Swr1 and Isw1,

which is explainable because INO80 is essential in

H2A.Z correct deposition (Lindstrom et al., 2011)

and may catalyze the removal of unacetylated H2A.Z.

We, therefore, speculated that the presence of NuA4

as histone acetyltransferase and INO80 as remover

of unacetylated H2A.Z at promoter in repressing

pathway deﬁnes the low level of gene expression;

and that in the pathway, INO80 help establishing the

periodicity of nucleosome organization.

4 CONCLUSIONS

We present here the reconstruction of interaction net-

work among various PTMs and transcriptional regu-

lators, with focus on their relationships with the dy-

namic nucleosome organization at promoter. Having

a large number of relationships correctly recovered,

the network features the regulatory processes that

show their presence by changing nucleosome orga-

nization at promoter, e.g., gene repression, postinitia-

tion regulation and PIC assembly/disassembly. Also,

we provide evidences for the hypothesis that nucleo-

some dynamics at promoter is regulated by extrinsic

factors, such as CRCs and GTFs. With these results,

the reliability of our method is proved, in addition to

the proposed valid learning procedure; hence, it can

be used to build networks of other factors.

REFERENCES

Cui, P., Zhanga, L., Lina, Q., Dinga, F., Xina, C., Fanga, X.,

Hua, S., and Yua, J. (2010). A novel mechanism of

epigenetic regulation: Nucleosome-space occupancy.

391(1):884–89.

Friedman, N., Linial, M., Nachman, I., and Peer, D. (2000).

Using bayesian networks to analyze expression data.

7(3-4):601–20.

Heckerman, D., Geiger, D., and Chickering, D. (1995).

Learning bayesian networks: The combination of

knowledge and statistical data. 20:197–243.

Jensen, F. and Nielsen, T. (2007). Bayesian Networks

and Decision Graphs. Springer-Verlag, New York, 2nd

edition.

Kaplan, N., Moore, I., Fondufe-Mittendorf, Y., Gossett, A.,

Tillo, D., Field, Y., Hughes, T., Lieb, J., Widom, J.,

and Segal, E. (2010). Nucleosome sequence pref-

erences inﬂuence in vivo nucleosome organization.

17:918–20.

Li, B., Carey, M., and Workman, J. (2007). The role of

chromatin during transcription. 128(4):707–19.

Lindstrom, K., Vary, J., Parthun, M., Delrow, J., and

Tsukiyama, T. (2006). Isw1 functions in parallel with

the nua4 and swr1 complexes in stress-induced gene

repression. 26(16):6117–29.

Lindstrom, K., Vary, J., Parthun, M., Delrow, J., and

Tsukiyama, T. (2011). Global regulation of h2a.z lo-

calization by the ino80 chromatin-remodeling enzyme

is essential for genome integrity. 144(2):200–13.

Luger, K., Mader, A., Richmond, R., Sargent, D., and Rich-

mond, T. (1997). Crystal structure of the nucleosome

core particle at 2.8 a resolution. 389:251–60.

Luksend, E., Ranjan, A., FitzGerald, P., Mizuguchi, G.,

Huang, Y., Wei, D., and Wu, C. (2010). Step-

wise histone replacement by swr1 requires dual ac-

tivation with histone h2a.z and canonical nucleosome.

143(5):725–36.

Morse, M. (2007). Transcription factor access to promoter

elements. 102(3):560–70.

Pinskaya, M., Nair, A., Clynes, D., Morillon, A., and Mel-

lor, J. (2009). Nucleosome remodeling and transcrip-

tional repression are distinct functions of isw1 in sac-

charomyces cerevisiae. 29(9):241930.

Pokholok, D., Harbison, C., Levine, S., Cole, M., Han-

nett, N., Lee, T., Bell, G., Walker, K., Rolfe, P.,

Herbolsheimer, E., Zeitlinger, J., Lewitter, F., Gif-

ford, D., and Young, R. (2005). Genome-wide map

of nucleosome acetylation and methylation in yeast.

122(8):517–27.

Samorodnitsky, E. and Pugh, B. (2010). Genome-wide

modeling of transcription preinitiation complex disas-

sembly mechanisms using chip-chip data.

Venters, B. and Pugh, B. (2009). A canonical promoter or-

ganization of the transcription machinery and its reg-

ulators in the saccharomyces genome. 19(3):360–71.

Wan, J., Lin, J., Zack, D., and Qian, J. (2009). Relating

periodicity of nucleosome organization and gene reg-

ulation. 25(41):1782–8.

Yu, H., S, S. Z., Zhou, B., Xue, H., and Han, J. (2008).

Inferring causal relationships among different histone

modiﬁcations and gene expression. 18(8):1314–24.

BIOINFORMATICS 2012 - International Conference on Bioinformatics Models, Methods and Algorithms

302