GENOMIC ANALYSIS OF NEURON-RESTRICTIVE SILENCER FACTOR

ACTIVITY IN NEURONAL AND NON-NEURONAL HUMAN CELL LINES

A DISSERTATION

SUBMITTED TO THE PROGRAM IN GENETICS

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Evonne C. Leeper

June 2010

http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/fn296jx7207

© 2010 by Evonne Chen Leeper. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.

ii

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Richard Myers, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Gregory Barsh

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Anne Brunet

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Margaret Fuller

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

iv

Abstract

The neuron-restrictive silencer factor/RE-1 silencing transcription factor

(NRSF/REST) is thought to be a negative regulator of neuronal genes in non-neuronal

cells. However, evidence of its continued expression and activity in neurons suggests

that NRSF may play other roles. A complete knowledge of NRSF target genes in

neuronal and non-neuronal cells is the first step to understanding its functions. Using

chromatin immunoprecipitation and quantitative PCR, I experimentally tested the

occupancy of NRSF in living non-neuronal cells on 113 candidate binding sites

predicted on the basis of conservation across the human, mouse, and dog genomes.

These tests helped to further refine the prediction algorithm and identified a number of

NRSF-bound regulatory microRNAs that may work in a feedforward loop to

downregulate NRSF and its co-repressor, CoREST. I next focused on understanding

NRSF recruitment in neuron-derived versus non-neuronal cell lines, using chromatin

immunoprecipitation paired with ultrahigh-throughput sequencing (ChIP-seq) to get a

direct, genome-wide picture of NRSF binding in human neuron-derived and non-

neuronal cell lines. I found a large overlap in the NRSF binding pattern between the

two cell types, particularly in binding sites found to be strongly or commonly bound.

There is a subset of strong sites bound in all cell types, and weaker sites that are more

likely to be cell-type specific. These common sites contain the canonical NRSE while

the cell line unique sites do not. Finally, I used another ultrahigh-throughput

sequencing based method to catalog and quantify all mRNA transcripts in each of the

cell lines (RNA-seq) to add target gene expression to the analysis of NRSF function.

Common target genes were more likely to be highly expressed in the neuron-derived

v

cell line than in non-neuronal cell lines despite NRSF binding in both. I also found

that the neuron-specific binding sites were primarily located in exons and promoters,

while common or non-neuronal specific binding sites were primarily located in introns

and intergenic regions. Differences in binding strength and target gene expression

levels suggest that NRSF has different binding mechanisms and functions in neuron-

derived and non-neuronal human cell lines.

vi

Acknowledgements

I would like to thank my advisor Rick and all the members of the Myers lab for their

help with my project and for teaching me so much during my time at Stanford. I would

like to thank my committee members Greg Barsh, Ann Brunet, and Minx Fuller for

their guidance over the years. I would like to thank my parents and my sister for their

love, support, and encouragement in all my academic endeavors. Finally, I would like

to thank my husband Josh, whose love, empathy, reassurance, and assistance in all the

other things of life made it possible for me to come this far.

vii

Table of Contents

ABSTRACT ..................................................................................................................................... IV

ACKNOWLEDGEMENTS ............................................................................................................. VI

LIST OF TABLES ......................................................................................................................... VIII

LIST OF FIGURES ....................................................................................................................... VIII

METHODS....................................................................................................................................... IX

CHAPTER 1: INTRODUCTION ..................................................................................................... 1

CHAPTER 2: COMPARATIVE GENOMICS MODELING AND EXPERIMENTAL

VALIDATION OF NRSES ............................................................................................................. 28

CONSERVATION-BASED GENOMIC COMPUTATIONAL PREDICTION OF NRSES ................................ 29

CHROMATIN IMMUNOPRECIPITATION ANALYSIS OF PREDICTED NRSES ........................................ 33

ANALYSIS OF NRSE2 MATCHES ASSOCIATED WITH MICRORNAS ................................................... 35

CHAPTER 3: COMPARISON OF GENOME-WIDE NRSF BINDING IN NEURON-DERIVED

AND NON-NEURONAL CELLS ................................................................................................... 38

CHIP-SEQ ANALYSIS OF HUMAN NEURON-DERIVED AND NON-NEURONAL CELL LINES ................... 40

LIBRARY COMPARISON AND PEAK CALLING HIGHLIGHT SIMILARITIES AND DIFFERENCES

BETWEEN CELL LINES ......................................................................................................................... 41

ASSEMBLY OF GENE COHORTS ASSOCIATED WITH CHIP-SEQ PEAKS .............................................. 45

MOTIF ANALYSIS OF CHIP-SEQ PEAKS IN EACH CELL LINE ............................................................. 50

CHAPTER 4: EXPRESSION ANALYSIS OF POTENTIAL NRSF TARGET GENES IN

NEURON-DERIVED AND NON-NEURONAL CELLS ............................................................... 52

RNA-SEQ ANALYSIS OF HUMAN NEURON-DERIVED AND NON-NEURONAL CELL LINES .................... 53

CORRELATION OF PEAK STRENGTH AND EXPRESSION LEVEL OF ASSOCIATED GENES IN EACH CELL

LINE ..................................................................................................................................................... 55

EXPRESSION OF GENES IN CELL LINE UNIQUE, COMMON, NEURON-DERIVED, AND NON-NEURONAL

COHORTS ACROSS ALL CELL LINES .................................................................................................... 59

ANALYSIS OF NRSF BINDING AND TARGET GENE EXPRESSION INCORPORATING CHIP-SEQ AND

RNA-SEQ DATA FROM OTHER NON-NEURONAL CELL LINES............................................................. 64

CHAPTER 5: DISCUSSION AND FUTURE DIRECTIONS ........................................................ 75

WORKS CITED ............................................................................................................................. 82

viii

List of Tables

Table 1: MicroRNAs with associated NRSE2 matches are expressed in brain ............................. 36

Table 2: ChIP-seq library comparisons .......................................................................................... 43

Table 3: Numbers of ChIP-seq reads and MACS peaks for all libraries ........................................ 44

Table 4: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts .......................... 47

Table 5: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts .......................... 72

Table 6: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts .......................... 74

List of Figures

Figure 1: Computational approach to NRSE PSFM refinement .................................................... 31

Figure 2: Different seed motifs converge following motif refinement .......................................... 32

Figure 3: Selection of a threshold for NRSE2 and correlation of score with repression................ 32

Figure 4: Validation of anti-NRSF antibody for use in ChIP ......................................................... 34

Figure 5: Quantitative analysis of NRSF ChIP .............................................................................. 35

Figure 6: NRSF gene regulatory model ......................................................................................... 37

Figure 7: MEME motif analysis of NRSF peak sequences ............................................................ 51

Figure 8: Calculation of RPKM ..................................................................................................... 55

Figure 9: Correlation between peak strength and expression of associated gene .......................... 57

Figure 10: Expression of commonly bound, cell line uniquely bound, neuron-derived

commonly and uniquely bound, and non-neuronal commonly and uniquely bound genes ........... 62

Figure 11: Expression of NRSF and co-repressors in all cell lines ................................................ 64

Figure 12: Location distribution of common binding sites, expression and GO analysis of

common gene cohort ...................................................................................................................... 68

Figure 13: Location distribution of peaks unique to each cell line ................................................ 69

Figure 14: Expression of genes associated with peaks unique to each cell line ............................. 70

Figure 15: Location distribution of peaks unique and common to neuron-derived and non-

neuronal cell lines ........................................................................................................................... 73

Figure 16: Expression of genes associated with peaks unique and common to neuron-derived

and non-neuronal cell lines............................................................................................................. 73

Figure 17: Strength of peaks common to all cell lines ................................................................... 78

ix

Methods

Cell culture

Culture conditions were as follows: Jurkat cells were grown in Advanced RPMI 1640

(GIBCO Invitrogen Cell Culture) supplemented with 15% fetal bovine serum, 100

U/mL of penicillin- streptomycin, and 1 × Glutamax (GIBCO Invitrogen Cell Culture)

at 37°C with 5% CO2. BE(2)-C cells were grown in a 1:1 mixture of MEM and F12

(GIBCO Invitrogen Cell Culture) supplemented with 10% fetal bovine serum, 100

U/mL of penicillin- streptomycin, and 1 × Glutamax (GIBCO Invitrogen Cell Culture)

at 37°C with 5% CO2. HTB-11 and U-87 cells were grown in MEM (GIBCO

Invitrogen Cell Culture) supplemented with 10% fetal bovine serum, 100 U/mL of

penicillin- streptomycin, and 1 × Glutamax (GIBCO Invitrogen Cell Culture) at 37°C

with 5% CO2. PANC-1 cells were grown in Dulbecco’s Modified Eagle’s Medium

(GIBCO Invitrogen Cell Culture) supplemented with 10% fetal bovine serum, 100

U/mL of penicillin- streptomycin, and 2 × Glutamax (GIBCO Invitrogen Cell Culture)

at 37°C with 5% CO2. PFSK-1 cells were grown in RPMI 1640 (GIBCO Invitrogen

Cell Culture) supplemented with 1× HEPES, 10% fetal bovine serum, 100 U/mL of

penicillin- streptomycin, and 1 × Glutamax (GIBCO Invitrogen Cell Culture) at 37°C

with 5% CO2.

Chromatin immunoprecipitation

This protocol was adapted from the laboratory of Peggy Farnham

(http://mcardle.oncology.wisc.edu/farnham/protocols). We cross-linked the cells by

adding formaldehyde to a final concentration of 1% for 10 min. Cross-linking was

stopped by adding glycine to a final concentration of 0.125 M. Then, we collected 2 ×

107 cells per IP and washed once with 1× phosphate-buffered saline (PBS). We

resuspended the cells in lysis buffer (5 mM 1,4-piperazine-bis-[ethanesulphonic acid],

at pH8.0, 85 mM KCl, 0.5% NP-40, Protease Inhibitor Cocktail [Roche]) and

centrifuged to collect the crude nuclear preparation. We resuspended the crude

nuclear preparation in RIPA buffer (1 × PBS, 1% NP-40, 0.5% sodium deoxycholate,

0.1% sodium dodecyl sulfate [SDS], Protease Inhibitor Cocktail) and sonicated at

power output 5–6 with the Sonics Vibra-Cell VC130 (Sonics) four times for 30 sec

each on ice to produce an average DNA fragment size of 500 bp. We centrifuged the

chromatin solution at 4°C for 15 min at 20,000 rcf. Sonicated chromatin was

incubated with NRSF mouse monoclonal antibody (12C11; Chen et al. 1998) coupled

to sheep anti-mouse IgG magnetic beads (Dynabeads M-280, Invitrogen). After bead

pelleting, the supernatant was retained as mock IP DNA for use in quantitative PCR.

The magnetic beads were washed five times with wash buffer (100 mM Tris, 500 mM

LiCl, 1% NP-40, 1% Deoxycholate) and washed once with TE (10 mM Tris at pH 8.0,

1 mM EDTA). After washing, the bound DNA was eluted by heating the beads to

65°C in elution buffer (0.1 M NaHCO3 and 1% SDS). The eluted DNA and mock IP

x

DNA were incubated at 65°C for 12 h more to reverse the cross-links. Then, we

extracted with phenol chloroform and back extracted the organic phase once. We

concentrated the DNA in the aqueous phase using the QIAquick PCR Purification Kit

(Qiagen), substituting 3 volumes of Qiagen Buffer PM for 5 volumes of Qiagen Buffer

PB.

Quantitative PCR

We used Primer3 software to design primers by inputting 500 bp of upstream genomic

sequence and 500 bp downstream of each predicted NRSE. Each primer pair was

required to flank the NRSE. We performed real-time PCR to quantitate the absolute

amount of enriched DNA for each NRSE (amplicon size range between 60 and 217

bp, average size of 79 bp). Each reaction contained 3.5 mM MgCl2, 0.125 mM

dNTPs, 0.5 μM forward primer, 0.5 μM reverse primer, 0.1 × Sybr Green (Molecular

Probes Invitrogen Detection Technologies), 1 U Stoffel fragment (Applied

Biosystems), and template DNA in a final volume of 20 μL. For each amplicon, we

measured a standard curve of 50 ng, 5 ng, 500 pg, and 50 pg mock IP DNA in addition

to our replicate ChIP DNA samples. We measured product accumulation for 40

cycles on the Bio-Rad Icycler and calculated the threshold cycle for each dilution of

the standard curve. We then performed a linear regression to fit the threshold cycle

from our ChIP DNA sample to this standard curve and divided that result by the

amplicon size to measure the absolute number of genomic equivalents of that NRSE in

the pool of ChIP DNA. We measured the levels of five random nongenic,

nonconserved regions in each ChIP DNA preparation to normalize for any variation in

absolute quantities of DNA in each prep.

Library preparation for Solexa/Illumina sequencing

Each Solexa/Illumina library was prepared from 4 pooled individual IPs which were

dried down and resuspended to 33 μL with ddH2O or 200 ng of reverse crosslinked

total chromatin from each cell line to create a control library. All reagents were from

a Genomic DNA Sample Prep Kit (Illumina). End repair was performed by

incubating a mix of the IPed or control DNA, 5 μL 10X end-repair buffer, 5 μL 1

mg/mL BSA, 2 μL 10 mM dNTP mix, and 5 μL 3 U/μL T4 DNA polymerase at 20°C

for 15 min. 1 μL of 5 U/μL Klenow DNA polymerase was added followed by a

second incubation at 20°C for 15 min, and the DNA fragments were purified using the

QIAquick PCR Purification Kit (Qiagen) and eluted into 35 μL of EB buffer. 5’

phosphorylation of the fragments was accomplished by mixing the blunt end DNA

with 5 μL each of 10X T4 PNK buffer, 10 mM ATP, and 10 U/μL T4 PNK and

incubating at 37°C for 30 min. Resulting fragments were purified using the QIAquick

PCR Purification Kit and eluted into 32 μL of EB buffer. 3’ A addition was

xi

performed by mixing the fragments with 5 μL 10X Klenow buffer, 10 μL 1 mM

dATP, and 3 μL 5 U/μL Klenow fragment (3’ to 5’ exo minus) and incubating at 37°C

for 30 min. Resulting fragments were purified using the MinElute PCR Purification

Kit (Qiagen) and eluted into 10 μL of EB buffer. Adapter ligation was performed by

mixing the DNA fragments with 25 μL 2X DNA ligase buffer, 10 μL 1:10 diluted

adapter oligo mix, and 5 μL 1 U/μL DNA ligase and incubating at 20°C for 15 min.

Resulting fragments were purified using the QIAquick PCR Purification Kit and

eluted into 30 μL of EB buffer. PCR preamplification of adapter-modified fragments

[30 sec at 98°C; (10 sec at 98°C, 30 sec at 65°C, 30 sec at 72°C) x 25 cycles; 5 min at

72°C] was performed using 23 μL of the DNA fragments, 25 μL Phusion DNA

polymerase, and 1 μL each of PCR primer 1.1 and PCR primer 2.1. Resulting

fragments were purified using the QIAquick PCR Purification Kit and eluted into 30

μL of EB buffer. The size selection of this library was performed by gel

electrophoresis and subsequent excision and purification of DNA (QIAex II, Qiagen)

in the ~150- to 300-bp range. Finally, a second PCR amplification [30 sec at 98°C;

(10 sec at 98°C, 30 sec at 65°C, 30 sec at 72°C) x 18 cycles; 5 min at 72°C] was

performed using 1 μL of the size selected DNA fragments. Resulting fragments were

purified using the QIAquick PCR Purification Kit and eluted into 30 μL of EB buffer.

RNA preparation

For each 2 × 107

cells, the cells were washed twice with room temperature PBS. After

discarding the PBS, 2 mL RLT buffer with a 1:100 concentration of β-

mercaptoethanol was added to lyse the cells. The lysed cells were scraped from the

plate and sheared through a 20-guage needle on a 5 mL syringe 20 times. 2 times the

lysate volume of RNase free water and 65 μL Proteinase K (Qiagen) was added to the

lysate, and the mixture was incubated at 55°C for 20 min before being spun at 3600

RPM in a swinging bucket table top centrifuge for 5 min. Supernatent was collected

and half of its volume of 100% ethanol was added and the mixture shaken by hand.

The RNA was purified using a modified RNeasy Midi Kit (Qiagen) protocol. After

the sample was loaded onto the RNeasy midi column, it was washed with 2 mL RW1

buffer. Then 20 μL DNase I in 140 μL RPP buffer was added, and the column was

incubated for 15 min at room temperature. 2 mL RW1 was added, followed by a 5

min room temperature incubation and a 5 min spin at 3600 RPM. Two additions of

2.5 mL RPE buffer were followed by a 2 min 3600 RPM spin after the first addition

and a 5 min 4200 RPM spin after the second. RNA was eluted with two serial

additions of 250 μL EB buffer incubated on the column at room temperature for 5 min

and spun for 3 min at 4200 RPM.

xii

cDNA preparation

Total RNA (75 μg) was subjected to two rounds of hybridization to magnetic

Dynabeads Oligo(dT)25 beads (Invitrogen) according to the manufacturer’s protocol.

100 ng of the resulting mRNA was then used as template for cDNA synthesis. The

mRNA was first fragmented by addition of 5× fragmentation buffer (200 mM Tris

acetate, pH 8.2, 500 mM potassium acetate, 150 mM magnesium acetate) and heating

at 94°C for 2 min 30 s in a thermocycler and was then transferred to ice and run over a

Sephadex-G50 column (USA Scientific) to remove the fragmentation ions. 3 μg

random hexamers were added to prime first-strand reverse transcription according to

the manufacturer’s protocol (Invitrogen cDNA SuperScript double strand cDNA

synthesis kit). After the first strand was synthesized, a custom second strand synthesis

buffer (Illumina) was added, and dNTPs, RNase H and Escherichia coli polymerase I

were added to nick translate the second-strand synthesis for 2.5 h at 16°C. The

reaction was then cleaned up on a QIAquick PCR column (Qiagen) and eluted in 30 μl

EB buffer (Qiagen).

1

Chapter 1: Introduction

Transcriptional regulation is fundamental to the creation and maintenance of

different cell types in multicellular organisms. Differentiation relies on temporal and

tissue-specific programs of gene expression. This is achieved through recognition of

DNA motifs, often conserved in gene promoters, by protein families which are

expressed in a tissue-specific, developmental, temporal, or stimulus-specific

manner.1,2

These DNA binding proteins include helix-turn-helix, zinc finger, leucine

zipper, and helix-loop-helix proteins. Each category has its own mechanism for

interaction with DNA. In eukaryotes, a single DNA binding protein can bind several

binding sites having limited sequence similarity. In addition, different proteins can

bind the same DNA site with similar affinity, leading to potential competition for the

same site between proteins with different functions or complexing of several

polypeptides to alter binding specificity.3 Binding of protein regulators to inducible

regulatory sequences can be affected by factors such as heat shock, viral infection,

growth factors, steroids, and membrane depolarization. Protein binding of temporal

and tissue-specific regulatory sequences may be controlled by a combination of

restricted expression of protein regulators, accessibility of DNA binding site, and

tissue-specific or temporal activation of ubiquitously expressed protein regulators.

Evidence suggests that a gene expressed in several different cell types is most likely

regulated by different transcription factors acting on separate or common regulatory

sequences in each cell type.1

Gene regulation can be positive or negative. Positive regulation is

characterized by the interaction between basal transcription machinery bound to the

2

promoter and transcription factors bound to enhancers that lead to increased

expression of the associated gene. Enhancers have been shown to act on cis-linked

promoters over long distances and in a position-independent and orientation-

independent manner.1 Many enhancers, such as the GC box that is recognized by Sp1,

are common to the majority of promoters. Co-activators, such as CREB-binding

protein, which do not directly bind DNA, associate with DNA-bound positive

transcription factors to modify chromatin factors that repress expression, link to basal

transcriptional machinery, or make covalent modifications to alter components of the

transcriptional machinery.4,5

In negative regulation, transcription factors bound to

silencer elements can block binding of basal transcriptional machinery to the promoter

or prevent binding or function of activator proteins. In some cases repressors and

activators help each other bind DNA, and the repressor function is usually dominant.

Repressor proteins have been shown to recruit co-repressors that either ensure a

condensed chromatin structure or interfere with binding or function of transcription

machinery at the promoter.2,4,6

While many co-activators interact with histone

acetyltransferase, many co-repressors function via Sin3-histone deacetylase (HDAC),

so the ultimate repressive mechanism is identical between most repressors.4 Like

enhancers, repressors are able to function over large distances in a position- and

orientation-independent way, but the context of the binding site within the promoter

may affect the mechanism of repression. Exon-located binding sites can mediate

repression before transcription or as an RNA-based element.5 Most transcription

factors can function as either repressors or activators, depending on the transcription

factor concentration and the nature of the binding site and flanking sequences, which

3

can affect conformation and recruitment of different cofactors.5-7

The majority of cell

type or tissue-specific transcription factors that have been studied have been shown to

function as positive regulators. There is a bias towards thinking of eukaryotic gene

regulation in terms of activation because of this. Also, only about seven percent of the

DNA in the large eukaryotic genome is transcribed into RNA. While it may be

possible that a number of repressors are bound in the non-transcribed regions, their

influence was overlooked or underestimated because methods of detection focus on

RNA. However, many different repressors functioning in a wide range of biological

contexts have been identified.6,7

More recently, negative regulation was discovered to

have a significant role in the control of neuronal gene expression.8

The importance of negative regulation in neuronal genes was uncovered with

the discovery of a silencer element, called the neural-restrictive silencer element

(NRSE)9 or repressor element 1 (RE1)

10, upstream of a few neuron-specific genes.

Two laboratories used promoter-reporter fusion vectors transfected into HeLa, L6 rat

muscle, and PC12 rat pheochromocytoma cells to perform deletion analysis on the

upstream regions of the rat type II sodium channel gene (NaChII)10

and the superior

cervical ganglion-10 protein gene (SCG10)9, a neuronal marker. This identified a

silencer element that repressed expression in the non-neuronal lines, but not in PC12

cells. This element was found to be position- and orientation-independent, and able to

repress transcription from a heterologous promoter.8 Electrophoretic mobility shift

assays (EMSA) using extracts from HeLa, L6, 10T1/2 mouse embryo, 3T3 mouse

fibroblast, PC12, MAH rat adrenomedullary, and SY5Y human neuroblastoma cells

identified a protein binding to the 21 to 28 base pair sequence only in non-neuronal

4

cells. Mutation of the silencer element that disrupted binding also alleviated

repression of a reporter gene. Footprinting using extracts from L6 and 3T3 cells also

revealed the 28 base pair binding site as a protected, protein-bound region. The gene

encoding the transcription factor that recognizes NRSE/RE1, called the neuron-

restrictive silencer factor (NRSF)11

or the RE1-silencing transcription factor (REST)12

was soon cloned. The NRSF/REST cDNA was isolated by screening HeLa, B cell,

and T cell expression libraries for proteins that could bind an NRSE/RE1-containing

oligonucleotide in an EMSA.11,13

Separately, the same protein was identified in a

yeast one-hybrid screen of HeLa cDNAs. NRSF/REST was found to be a 116 kDa

protein with eight zinc fingers near the amino terminus and one zinc finger near the

carboxyl terminus, which put it in the GLI-Kruppel family of transcription factors.

Adjacent to the zinc finger DNA binding domain is a region rich in basic amino acids

followed by six proline-rich repeats.12,14

Western blotting and in situ hybridization

indicated high NRSF/REST expression in non-neuronal cells, including neuronal

precursors and glial cells, and very low expression in neurons. Cotransfection of

NRSF/REST cDNA and NRSE/RE1-containing reporter constructs into PC12 cells,

which lack endogenous NRSF/REST, demonstrated the repressive function of

NRSF/REST protein and the necessity of its binding to NRSE/RE1 to carry out

repression. Consensus NRSE/RE1 sequences were indentified in 18 neuron-specific

genes, and four of these were shown to be recognized by NRSF/REST, leading to

suggestion of NRSF/REST’s role as a master neuronal regulator.11

Most studies of NRSF/REST have focused primarily on its role as a repressor

of neuronal genes in non-neuronal cells. Transient transfection experiments to

5

characterize the regulatory regions of several neuronal genes identified repression by

NRSF and NRSE. Transfection of reporters driven by deletion constructs of the m4

muscarinic acetylcholine receptor gene regulatory region into L6, PC12, NG108-15 rat

neuroblastoma and mouse glioma hybrid, CHO hamster ovary, and 3T3 cells defined a

region that included an NRSE which repressed the reporter in non-neuronal lines, but

not in neuronal lines. Deletion or mutation of the identified NRSE abolished this

repression. EMSA using extracts from the various cell lines and the fragment

containing the NRSE found a protein only present in non-neuronal lines that

specifically binds the NRSE within this promoter fragment. Cotransfection of an

NRSF expression vector and a reporter driven by the NRSE-containing fragment into

NG108-15 cells, which lack endogenous NRSF-mediated repression, showed that the

exogenous NRSF expression was able to repress the reporter, but to a lesser extent

than it could a reporter driven by the NRSE-containing region from the rat type II

sodium channel gene. This implies that NRSF contributes to repression, but that the

makeup of each promoter and the proteins bound to it can change the level of

repression.15,16

Similar deletion analyses of regulatory regions connected to the

neuron-glia cell adhesion molecule gene (Ng-CAM)17

, the β2-subunit of the neuronal

nicotinic acetylcholine receptor gene18

, and the synapsin I gene19

done in various

mouse, human, hamster, and rat neuronal and non-neuronal cell lines also identified

NRSE-containing regions that mediated non-neuronal-specific repression of a reporter

gene, along with other regulatory regions that contributed to expression level. EMSA

experiments with the NRSE-containing region from Ng-CAM identified much more

protein binding in 3T3 cells than in N2A mouse neuroblastoma cells.17

Transgenic

6

mice incorporating a β-gal reporter driven by the neuronal nicotinic acetylcholine

receptor β2-subunit promoter showed neuron-specific reporter expression.18

Expression of all of these neuronal genes is restricted to neurons primarily by NRSF-

mediated repression in non-neurons, but their expression level is further adjusted by

other regulatory elements.

Biological activity of NRSF has been examined through the use of model

promoters and fusion proteins. A grp78 promoter-driven reporter with the synapsin I

NRSE inserted 2.3 kb downstream of the transcription start site cotransfected into

NS20Y mouse neuroblastoma cells with an NRSF expression vector revealed NRSF-

mediated repression despite distance. Repression of a similar reporter construct that

also included a strong SV40 enhancer was also seen. NRSF can repress promoters or

block strong enhancers regardless of distance or orientation. A GST fusion protein

that included the NRSF N-terminal zinc finger cluster was able to bind NRSEs derived

from synapsin I and SCG10 in an EMSA, while a fusion protein including the NRSF

C-terminal zinc finger could not.20

Expression of NRSF deletion constructs in PC12

cells transfected with a NaChII promoter-driven reporter showed that the N-terminal

zinc finger cluster, the DNA binding domain, does not cause repression when

expressed alone. Deletion of the proline motif alone did not affect repression, but

deletion of either the N-terminus of the C-terminus weakened repression of the

reporter.21

These repressor domains were further tested by creation of GAL4 fusion

proteins and reporters with UAS binding sites cotransfected into PC12, NS20Y, and

Neuro2a mouse neuroblastoma cells. Full-length NRSF, N-terminus amino acids 43

to 83, and C-terminus amino acids 989 to 1,097 could all cause significant repression.

7

The two termini could repress independently and did not seem to have an additive

effect. Mutation of the C-terminal zinc finger abolished this repression. At a distance

of 1,105 base pairs from the promoter, both termini were still able to cause repression,

but N-terminus-mediated repression was slightly stronger.20-22

These studies

characterized the NRSF DNA binding domain and its two individual repressor

domains at the protein termini.

Several co-repressors associating with either or both repressor domains of the

NRSF protein have been identified and characterized. Two yeast two-hybrid screens

using the NRSF N-terminus as bait identified mSin3A and mSin3B22,23

, while a screen

using the NRSF C-terminus as bait identified CoREST.24

Yeast SIN3 was also

identified as an important component to NRSF-mediated repression in a yeast

repression screen using NRSE-containing reporters and exogenously expressed NRSF

fragments. NRSF-mediated repression was lost in a sin3- yeast strain.25

Additional

yeast two-hybrid screens and in vitro binding assays specified that the C-terminal zinc

finger is important to the interaction between NRSF and CoREST and that binding of

NRSF to mSin3A occurs through the PAH2 domain of mSin3A.22-24

Co-

immunoprecipitation experiments identified in vivo interactions between NRSF and

both mSin3A and histone deacetylase (HDAC) in Neuro-2A mouse neuroblastoma

and C6 rat glioma cells, between NRSF and both CoREST and mSin3A in HEK-293

human embryonic kidney cells, between NRSF and mSin3B in 3T3 cells, and between

NRSF and CoREST in L6 cells.22-26

NRSF co-repressors may be the components of

an NRSF protein complex that functionally causes repression. A CoREST fusion

protein to the GAL4 DNA binding domain was able to repress a UAS-containing

8

reporter when cotransfected into both HEK-293 and PC12 cells, in the absence of

NRSF.24

Competition assays in which the NRSF C-terminus was overexpressed in

HEK-293 and PC12 cells to act as a dominant negative against CoREST or the NRSF

N-terminus was overexpressed in HEK-293 cells to act as a dominant negative against

mSin3A showed derepression of an NRSE-containing reporter in the presence of

either dominant negative, an effect rescued by additional overexpression of the

relevant co-repressor.23,24

Chromatin immunoprecipitation (ChIP) showed that

expression of the NRSF N-terminus dominant negative disrupted binding of both

mSin3A and HDAC at the NRSE of a reporter construct and resulted in higher levels

of histone acetylation.25

Expression of NRSF lacking a nuclear localization signal

resulted in NRSF, mSin3A, and CoREST colocalizing in cytoplasmic aggregates,

indicating that the proteins do not require DNA binding to interact. In situ

hybridization of mouse embryos showed ubiquitous NRSF and mSin3A expression,

but restriction of CoREST expression to head mesenchyme at E8.5. At E11.5, both

co-repressors were ubiquitously expressed, implying that both co-repressors may not

be needed simultaneously for repression and that the type of repression may be varied

through use of different co-repressor complexes.23

Treatment of cells with trichostatin

A (TSA), an HDAC inhibitor, resulted in derepression of NRSE-containing genes and

reporter constructs in an NRSE-dependent manner. This effect was stronger for

reporters repressed by the NRSF N-terminal domain than those repressed by the NRSF

C-terminal domain, indicating that HDAC plays a larger role in the NRSF N-terminal

repressor complex.22,25,26

Microinjection of antibodies against N-CoR, HDAC,

mSin3A, or NRSF all relieved repression of an NRSE-containing reporter in rat

9

fibroblast cells. ChIP in these same cells using antibodies against NRSF, N-CoR,

HDAC, and CoREST found all but CoREST bound to the SCG10 promoter.27

A

comparison of the co-repressors present at the SCG10 and NaChII genes in rat

fibroblast and the effects of TSA and 5’-aza-cytidine (5AzaC), which reverses DNA

methylation, on the expression of these genes showed that SCG10 repression depends

on NRSF with a complex of mSin3A, HDAC1-3, and N-CoR bound to its N-terminal

domain. This complex establishes a dynamic mode of regulation that can switch

between repression and activation. The NRSF C-terminal repression domain interacts

with CoREST and a wide variety of proteins involved in gene silencing through

chromatin condensation, including HDACs, MeCP2, histone methyltransferase, and

histone demethylase. This complex, along with DNA methylation, silences NaChII

and the nearby genes NaChIII and HoxD9, which do not contain NRSE/RE1

sequences.28,29

Different combinations of co-repressors can lead to different modes of

repression and differential expression of NRSF-regulated genes.

A mouse knockout of NRSF resulted in abnormal cell proliferation and

migration, widespread apoptosis, and 100% lethality by E11.5. Of six neuronal genes

investigated for derepression, including neuronal III β tubulin, SCG10, L1, synapsin I,

calbindin, and middle neurofilament, only βIII tubulin was derepressed.

Immunohistochemistry using TuJ1 antibody identified βIII tubulin upregulation in

several non-neuronal tissues. Although myotome cells were disorganized, they

retained normal expression of Myf5 and myogenin muscle regulatory genes, indicating

that lack of NRSF does not cause muscle cells to become neurons. The less dramatic

approach of mosaic inactivation by retroviral delivery of a dominant negative NRSF in

10

chicken embryos caused βIII tubulin, Ng-CAM, and SCG10 to become derepressed in

different patterns in non-neuronal cells and neuronal progenitors, indicating a

promoter-specific action of NRSF according to the co-repressor complex involved.

The NRSF knockout did not cause ectopic neurogenesis, implying that NRSF is not a

master regulator of neural induction, but is responsible for arranging appropriate gene

expression after determination of cell fate.30,31

Deletion and RNAi knockdown of

NRSF in mouse embryonic stem cells had no effect on expression or localization of

neural-specifying transcription factors or expression of brain-specific miRNAs,

indicating that NRSF does not have a role in repressing these neuronal factors in ES

cells. While some brain-expressed and NRSE-containing genes were seen to be

upregulated as a result of lost NRSF, only a small percentage of all NRSE-containing

genes were affected. Both NRSF-deficient and wildtype ES cells had similar

expression changes of pluripotency factors and neural genes in response to embryoid

body-forming conditions, so lack of NRSF did not cause loss of multi-lineage

potential.32

In contrast, NRSF+/- mouse ES cells and siRNA knockdown of NRSF in

mouse ES cells in a different study led to greatly reduced self-renewal measured in

alkaline phosphatase assays, and this could be rescued by additional exogenous NRSF

expression. Quantitative RT-PCR of the haploinsufficient cells showed increased

expression of markers for ectoderm, mesoderm, endoderm, and trophectoderm, much

like the level seen in embryoid bodies, ES cells grown in differentiation conditions.

Only one of the markers seen to be upregulated is a direct NRSF target, so NRSF

works directly and indirectly to maintain self-renewal in ES cells. Decreased

expression of self-renewal genes Oct4, Nanog, Sox2, Tbx3, and c-myc was also seen

11

in NRSF+/- cells. Also, the miRNAs found to be expressed or repressed was reversed

when comparing haploinsufficient and wildtype ES cells. Many of those upregulated

by reduced NRSF are thought to target self-renewal genes, and several had nearby

NRSEs seen to be occupied by NRSF in wildtype ES cells. miR124a, a target of

NRSF, and miR106a and miR106b, which are predicted to target NRSF, were all

upregulated in NRSF+/- cells, suggesting a possible double negative feedback loop.33

In another study, siRNA knockdown of NRSF in human mesenchymal stem cells

resulted in development of neurite-like structures, slowed growth, increased

expression of several neuronal genes, eventual expression of mature neuronal markers,

development of neuron-specific Nissl bodies, and development of functional voltage-

gated ion channels.34

While loss of NRSF in ES may cause a loss of pluripotency, it

specifically caused terminal neuronal differentiation in mesenchymal stem cells.

Two studies in which NRSF-bound genes were activated rather than repressed

suggest that their activation is adequate to induce neuronal differentiation. A fusion

protein was created linking the NRSF DNA binding domain to the strong viral

activator VP16, allowing strong activation of genes normally targeted by NRSF. A

vector encoding this fusion protein was transiently transfected into NT2 human

teratocarcinoma cells, which resemble committed neuronal progenitors and have a low

level of endogenous NRSF-mediated repression. RT-PCR analysis showed expression

of GluR, a neuronal differentiation marker, in NT2 cells only after transfection with

the NRSF-VP16 fusion or after treatment with retinoic acid (RA) to induce neuronal

differentiation. Upregulation of other neuronal, NRSE-containing genes and

downregulation of NeuroD3, a marker of immature neurons, was also seen as a

12

consequence of NRSF-VP16 expression. However, the cells did not terminally

differentiate into neurons despite increased expression of differentiation genes. It is

possible that the activation was not continued for a long enough period of time,

expression of other genes are required, the genes were not expressed in an appropriate

progression, or expression of stem cell markers was not suppressed.35

Stable

integration of doxycycline-inducible sequence encoding the NRSF-VP16 fusion into

mouse clonal neural stem cells caused expression of the neuronal differentiation

marker neuronal βtubulin and synaptic vesicle protein synaptotagmin I in addition to

development of neurite-like structures within 16 days of induction. NRSF-VP16

induction caused the neural stem cells to become sensitive to RA-treatment, leading to

differentiated cells that survive in the presence of mitotic inhibitors, express Tuj1 and

MAP2 neuronal differentiation markers, and form neurite-like structures. Induced

cells also demonstrated rapid, reversible, depolarization-dependent calcium influx

when depolarized with high potassium or glutamate, a physiological property of

neurons.36

Finally, computational approaches have been used to identify potential NRSF

binding sites on a genome-wide scale. A GenBank search using a composite NRSE

identified 22 genes, 17 of which are primarily neuronally expressed.37

Sixteen

neuronal and eight non-neuronal NRSEs were tested by EMSA for in vitro binding by

in vitro-translated human NRSF, resulting in 15 neuronal and five non-neuronal bound

NRSEs. Each of these was able to compete binding of in vitro-translated human

NRSF and NRSF from HeLa extracts to SCG10 NRSE. Thirteen of these NRSEs

were placed upstream of an SCG10 promoter-driven reporter and transfected into

13

mouse fibroblasts to test for functional repression, resulting in identification of five

new neuronal and two non-neuronal genes repressed by NRSF. These functional tests

revealed a 14 bp core sequence required for binding in the NRSE.37

The most

intensive search used a consensus NRSE built from 32 known NRSEs to search

ENSEMBL genome databases and discovered 1,892 human, 1,894 mouse, and 554

fugu NRSEs with 355, 358, and 416 of these falling within 10 kb 5’ of a gene and 593,

564, and 181 falling within genes respectively. 40% of the associated genes are

expressed in the nervous system, but many of them have no obvious neuronal function

or are required in both neuronal and non-neuronal cells. The most common NRSE

permutation were tested for NRSF binding using extracts from rat lung fibroblast, and

the only sequences that were not bound were those found in repetitive regions far from

any genes. A ChIP performed in glioma cells only identified the L1 cell adhesion

molecule (L1CAM) and synaptosomal-associated protein 25 (SNAP25) NRSE-

containing regions as enriched. When this was repeated in glioma cells with

additional exogenous NRSF expression, all known targets were bound, but L1CAM

and SNAP25 were still more highly enriched. Further computational analysis

identified two closely spaced NRSEs near each of these genes. Expression of NRSF

target genes in glioma cells transfected with additional NRSF or an NRSF dominant

negative was examined. SCG10 was silent in both wildtype and transfected cells.

Only SNAP25 was derepressed in the presence of dominant negative, and additional

NRSF had no effect on the targets. Within the same cell, different genes react

differently to the same NRSF expression level.38

Interestingly, NRSE sites were

14

found in both neuronal and non-neuronal genes, and some NRSF targets were found to

be neural transcription factors implying a wider role for NRSF.37,38

There is disagreement between studies of NRSF in neurons, but a great deal of

evidence has been found to show that NRSF function is not confined to repression in

non-neuronal cells. Despite the undetectable level of NRSF observed in some

neuronal cell lines, a few studies have shown NRSF expression in neuronal lines and

brain tissues. When full length human NRSF was used to screen a rat neuronal

progenitor-derived cDNA library, three different 5’ untranslated regions were found

associated with NRSF transcripts, and a clone predicting a truncated version of the

protein with only four zinc fingers was found in addition to the full length clone. A

northern blot found varying levels of NRSF mRNA in adult brains tissues such as

striatum, thalamus/hypothalamus, pons/medulla, hippocampus, cerebellum, midbrain,

septum, olfactory bulb, cerebral cortex, and colliculi as well as in non-neuronal tissues

such as testis, spleen, and muscle. In situ hybridization was able to specifically locate

NRSF expression in adult brain neurons of hippocampus, pons/medulla, and midbrain

as well as strong expression in non-neuronal brain cells. RNase protection assays and

RT-PCR identified two shorter alternatively spliced versions of NRSF expressed only

in neurons. After kainate-induced seizures, expression of NRSF and the two NRSF

isoforms was induced in hippocampus by four hours after injection and remained

elevated for at least 24 hours. In situ hybridization was able to locate this expression

to granular neurons of the dentate gyrus, pyramidal layers, cerebral cortex layers, and

piriform cortex.39

After finding that exogenous NRSF did bind to the NRSE

associated with m4 muscarinic acetylcholine receptor (m4 mAChR) and decrease

15

expression without completely silencing the gene in PC12 cells, one study investigated

the state of the m4 mAChR gene in four brain regions. Nuclei were isolated from

three regions where m4 mAChR is expressed: striatum, hippocampus, and cortex, and

one region where it is not expressed: cerebellum. Several DNase I hypersensitive sites

were found in the m4 mAChR gene in the tissues where it was expressed, matching

sites previously found in the PC12 expressing exogenous NRSF. Only one DNase I

hypersensitive site was found in the gene in cerebellum. ChIP found NRSF present at

the m4 mAChR NRSE in cortex but not in cerebellum. Interestingly, NRSF was not

found at the NRSE of the silent m4 mAChR gene in rat fibroblasts, and treatment of

these cells with HDAC inhibitor, methylation inhibitor, or an NRSF dominant

negative did not relieve the silencing, indicating that NRSF is not present or necessary

for m4 mAChR silencing in these non-neuronal cells. These data show that NRSF

may act as a regulator of actively transcribed genes in brain tissues and possibly in

neurons, and that NRSF is not required for maintenance of silencing at all NRSE-

containing genes in all cells.40

Many studies have examined NRSF activity during neuronal development.

While one group found high NRSF expression at embryonic stages that diminished but

continued into adult brain tissues39

, another found a gradual decrease in NRSF mRNA

in NB-OK-1 human neuroblastoma cells over the course of their neuronal

differentiation by staurosporine and forskolin treatment. Over the 17 day

differentiation, synapsin I mRNA level increased. Although NRSF and synapsin

expression levels varied between six human neuroblastoma cell lines, they always

maintained an inverse relationship.41

Another group examined mouse ES cells

16

allowed to form embryoid bodies and neuronally differentiated by RA treatment. At

the ES cell stage, immunostaining identified NRSF in nuclei, ChIP verified NRSF

binding at the calbindin NRSE, a reporter including a UAS site was silenced by both

NRSF-Gal4 and CoREST-Gal4 fusion proteins, and coIP showed HDAC in a complex

with CoREST. After differentiation, no more NRSF staining was observed, but

CoREST and mSin3A proteins were found to be present, and CoREST was still found

in complexes with HDAC implying continuing repression mediated by CoREST in the

absence of NRSF. Interestingly, nuclear run-on analysis of these cells at earlier stages

of differentiation found no decrease in NRSF transcription rate despite a very early

and dramatic decrease in NRSF protein concentration, suggesting posttranslational

downregulation. Proteasome inhibitor treatment could restore NRSF protein

concentration in ES cells and cells at neuronal progenitor stages, but not in postmitotic

cortical neurons. ChIP showed NRSF binding at NRSEs of several neuronal genes

through the progenitor stage, but not in postmitotic neurons. In cortical progenitors,

treatment with a NRSF dominant negative led to derepression of several known target

genes, indicating that NRSF still functions as a repressor of many neuronal genes at

this stage. Investigation of several NRSF target genes over the course of

differentiation found them to be released from repression at different times and their

eventual expression to be at different levels, indicating distinct regulatory mechanisms

and differential affinities for NRSF. Methylation was found at some NRSEs and CpG

sites in target genes distinct from the NRSE throughout neuronal differentiation. The

NRSF repressor complex is bound to the NRSE and other mCpG sites in ES cells, then

MeCP2 and co-repressors remained bound to the mCpG after NRSF has left the

17

NRSE. This study proposed a model for neuronal differentiation in which NRSF is

degraded to low levels during the switch from pluripotent cells to neural progenitors,

keeping the target chromatin inactive but poised for expression. At the switch from

neural progenitor to mature neuron, NRSF dissociates from the NRSE, NRSF

expression is downregulated by a repressor complex that includes many NRSF co-

repressors, and target genes are continually regulated by CoREST and MeCP2 binding

at a site distinct from the NRSE.42

This model was later modified to include a separate

mechanism for the differentiation of an adult stem cell into a neuron.29

This

modification incorporated the discovery of a small, non-coding double-stranded RNA

that includes the NRSE sequence and is bound by NRSF in adult stem cells, resulting

in the expression of neuronal genes and transition to neuronal cells. NRSF was found

to remain bound to the NRSE sites of the expressed neuronal genes, leading to the

proposal that the dsRNA transforms NRSF into an enhancer.43

Finally, a study using

ChIP-based cloning to identify NRSF target genes in mouse ES cells, embryonic

hippocampal neural stem (NS) cells, and mature hippocampus at also found that many

of the NRSF-bound targets are important for neuronal function and are lowly

expressed in ES and NS and highly expressed in mature neurons. While

immunostaining revealed a decrease in NRSF expression over ES cell differentiation,

NRSF was still present in NS cells and in specific neuron types in the hippocampus.

Of the 93 clones found using the ChIP cloning method, 89 were proximal to genes. 24

of the 38 annotated genes are nervous system specific. NRSF was found to be bound

near target genes highly expressed in hippocampus, suggesting that it may not be

functioning as a repressor in those cases.44

18

Studies of nervous system tissues and neuronal cells in which NRSF has been

perturbed seem to support a role for the regulator in neuronal differentiation.

Unilateral electroporation of stage 12-13 chick embryos with full-length mouse NRSF

cDNA in a retroviral vector led to constitutive NRSF expression throughout the

electroporated side of the spinal cord. Immunostaining showed down-regulation of

NRSF reporter genes N-tubulin and Ng-CAM when compared to the control side of

the embryo or an embryo electroporated with a negative control vector. To check for

appropriate neuronal morphology and connectivity, stage 13-14 chick embryos were

electroporated with an expression cassette designed to express both NRSF and GFP

fused to the microtubule-associated protein tau which would allow visualization of cell

body morphology and cell processes. Immunostaining showed high NRSF and GFP

expression in cells of the electroporated side of the embryo. NRSF overexpression did

not prevent attainment of neuronal morphology, but did cause axonal pathfinding

errors.45

In a second study of NRSF overexpression in neuronal cells, a PC12 line

with NRSF under control of a tet inducible promoter and a control line with only the

tet inducible promoter were created. In the control, western blotting showed no NRSF

expression, and the cells responded to nerve growth factor (NGF) treatment with an

increase in NaChII expression, leading to an increased inward sodium current

observed by whole-cell electrophysiology recording. These cells extended neurites of

a similar length and complexity to those found in other PC12 cell lines. When NRSF

expression was induced in the inducible line, NRSF was expressed and able to repress

a transiently transfected reporter in an NRSE-dependent manner. NRSF expression

was able to completely block induction of NaChII expression by NGF treatment,

19

resulting in a three-fold lower sodium current than NGF-treated cells with no NRSF

induction. Neurite length in NRSF induced, NGF treated cells was also three-fold

lower than in NGF treated cells without NRSF induction. Similar results were seen in

a primary culture of mouse cortical neurons infected with an NRSF-expressing virus,

suggesting that NRSF downregulation is important for both induction and

maintenance of neuronal phenotype.46,47

One group examined the effect of reduced

NRSF on neuronal cells by transfecting N18 mouse neuroblastoma cells with NRSF

siRNA, resulting in a significant knockdown of NRSF mRNA levels and decrease of

NRSF protein concentration. After 24 hours, there was a significant increase in length

and number of neurites. Interestingly, of four NRSE-containing genes involved in

neurite outgrowth analyzed by qPCR, L1 and Ulip1 had increased expression after

NRSF knockdown while Elmo2 and Ulip2 had decreased expression. This study

showed that NRSF downregulation leads to development of a neuronal phenotype as

expected, and NRSF has differing effects on transcription in neuronal cells.48

Several groups studying individual genes have identified that NRSF can

function as both a repressor and an activator of the nicotinic acetylcholine receptor β2-

subunit (nAChR), the cell adhesion molecule L1, corticotrophin releasing hormone

(CRH), and dynamin I, all genes important to neuronal function. In a study of NRSF

regulation of nAChR, an NRSE fused upstream of a ubiquitous SV40 promoter-driven

reporter and transfected into SK-N-Be human neuroblastoma, PC12, and 3T6 mouse

fibroblast cells was silenced in all three cell lines. However, when the SV40 promoter

was substituted with a minimal promoter, the NRSE mediated silencing in 3T6 cells

and enhancement in SK-N-Be and PC12 cells. A series of reporter constructs with

20

different spacing between the NRSE and TATA box of a minimal promoter showed

that the NRSE mediated enhancement when located less than 50 base pairs upstream

from the TATA box or anywhere in the 5’ UTR, but weak repression when further

upstream in neuroblastoma cells. The same constructs showed that NRSE always

mediates repression in fibroblasts.49

L1 regulation was examined using transgenic

mice which had LacZ reporters controlled by the L1 regulatory region with an intact

or deleted NRSE. As expected, deletion of the NRSE led to ectopic reporter

expression in many non-neuronal tissues during postnatal development. A dramatic

increase in reporter expression was also seen in neurons throughout the brain at birth,

glia surrounding nerve bundles in spiral ganglion of the ear, and olfactory ensheathing

cells. Surprisingly, NRSE deletion resulted in a reduction of reporter expression later

in postnatal development and in adult brain and nervous system structures. As the

NRSE is 10 kb away from the L1 promoter, close proximity did not seem to be

required for enhancer activity.50

CRH promoter-driven reporters which had intact or

mutated NRSEs behaved as expected when transiently transfected into L6, PC12, and

NG108-15 cells with or without an NRSF expression vector. However, when the

same reporters and expression vectors were transfected into NG108-15 cells which

had been differentiated into a more mature neuronal phenotype by forskolin and

IBMX treatment, significant upregulation of the reporter containing mutated NRSE

was seen in the presence of exogenous NRSF, suggestion that NRSF could function as

a CRH repressor via the NRSE and a CRH enhancer independently of the NRSE in

neuronal cells.51

Similar reporters driven by the dynamin I promoter with or without

the NRSE were transfected into mouse lung carcinoma and NS20Y mouse

21

neuroblastoma cells with or without an NRSF expression vector. In lung cells, the

reporter was silent even in the absence of NRSE, indicating that loss of NRSF

repressor activity is not sufficient to relieve repression of the dynamin I gene. In

NS20Y cells, the reporter with the intact NRSE was enhanced in the presence of

exogenous NRSF, indicating that NRSF activation of this gene occurs through NRSF

interaction with the NRSE.52

Clearly, genetic and cellular contexts affect NRSF

function, and may help make a distinction between different neuronal cell types as

well as the distinction between neuronal and non-neuronal cell types.

A few potential mechanisms for the activator function of NRSF have been

identified and characterized. Sequencing of 20-40 nucleotide RNAs extracted from

HCN-A94 adult rat hippocampal neural stem cells over the course of RA and

forskolin-induced neuronal differentiation identified sense and antisense NRSE RNAs

in the neuronal cell population, but only at very low levels in progenitors and absent in

the astrocyte population. Expression of both sense and antisense NRSE RNA from a

lentiviral vector in HCN-A94 cells caused the cells to extend long processes, create

large flat clusters, and express neuron-specific markers TUJ1, NF200, and calbindin.

No morphological changes were observed when either NRSE RNA was expressed

alone. Infection with the NRSE dsRNA decreased expression of reporters driven by

promoters taken from progenitor-specific, astocyte-specific, or oligodendrocyte-

specific genes, but increased expression of a reporter driven by the neuron-specific

TUJ1 promoter and several NRSE-containing genes. Experiments using NRSE-

containing reporters showed that their expression was increased by coinfection of

NRSE dsRNA in HCN-A94 progenitor culture, mouse neurosphere cultures, and

22

mouse primary neural stem cells taken from the ventricular zone, hippocampus, and

whole brain. This activation required both and intact NRSE near the target gene and

an intact NRSE in the dsRNA. ChIP and oligoIP showed that after introduction of

NRSE dsRNA, NRSF stayed bound to target NRSEs, fewer co-repressors were bound

to NRSF, and NRSF has a higher binding affinity for NRSE dsRNA than for NRSE

dsDNA. In mouse, this NRSE dsRNA is expressed only in regions where adult

neurogenesis is continuously occurring, and it could be binding directly to NRSF

bound at target genes or to an NRSF homodimer to cause activation of NRSF target

genes leading to neuronal differentiation.43

A posttranscriptional mechanism was

discovered in a study of the regulation of the mu opioid receptor (MOR), an NRSE-

containing gene. Exogenous NRSF expression in SHSY5Y and NMB human

neuroblastoma cells resulted in increased expression of a MOR-GFP fusion protein

and increased opioid-ligand binding activity by endogenous MOR, but decreased

MOR transcription. This posttranslational activation was not seen in PC12 cells

which have NRSF only in the nucleus, while NMB and rat and mouse primary

hippocampal neurons have NRSF both in the nucleus and cytoplasm. In order to

separate transcriptional and translational effects, MOR-reporter fusion mRNA

transcripts with intact or mutated NRSE were directly transfected into NMB cells

along with an NRSF expression vector, resulting in increased reporter expression from

only the transcripts with intact NRSEs. An EMSA verified NRSF binding to the intact

sense NRSE of the mRNA. After cotransfection of a MOR-reporter vector and and

NRSF expression vector into NMB cells, reporter transcripts and NRSF protein were

both found to be concentrated in the polysome fraction, suggesting that NRSF may

23

interact with the NRSE of the MOR transcript and promote its localization to a

ribosome complex resulting in enhanced translation.53

Two studies outline the

function of REST4, a neuron-specific splice variant of NRSF which includes the N-

terminal end of NRSF truncated such that the protein only includes five zinc fingers.

In a study of the cholinergic gene locus (CGL) in PC12 cells, it was found that protein

kinase A (PKA) activity increased expression of the CGL in and NRSE-dependent

manner, and this was not due to any change in NRSF protein concentration. EMSA

showed that NRSF was only bound to the CGL NRSE in PC12 which were PKA-

deficient, implying that PKA activity somehow prevents or disrupts NRSF DNA

binding. RT-PCR identified REST4 mRNA only in cells with PKA activity, but

western blotting of PC12 cells showed that REST4 is present at concentrations much

lower than NRSF. EMSA showed that REST4 did not bind to the NRSE, but was able

to block NRSF binding to the NRSE when extracts containing the two proteins were

mixed. REST4 and NRSF were co-immunoprecipitated, leading to a model where

PKA activity induces expression of REST4 which binds to NRSF, preventing it from

binding to NRSE and repressing CGL.54

Another study examined the effects of each

NRSF repression domain on the expression of the glucocorticoid response element

(GRE) by using each domain fused to the Gal4 DNA binding domain and GRE with a

UAS binding site. While full-length NRSF and the NRSF C-terminal domain

repressed expression, the NRSF N-terminal domain stimulated GRE expression. As

the NRSF N-terminal domain used is very similar to REST4, it was also tested for

enhancer ability in Neuro2A mouse neuroblastoma cells cotransfected with NRSE-

containing GRE vector, and REST4 was found to enhance expression.55

Overall, the

24

research of NRSF in neurons thus far indicates the continued expression and function

of NRSF in neurons and has begun to outline a much more complex role than the

originally proposed repression.

A more complete knowledge of NRSF binding and function in various cell

types could lead to a better understanding of its role in disease. NRSF has been

identified to function as both a tumor suppressor and an oncogene depending on

cellular context. While normal human bronchial epithelial cells express epithelial

markers, small cell lung cancer primary samples and cell lines have been seen to

express epithelial and neuronal markers, a hallmark of neuroendocrine tumors.

Further investigation of these cancer tissues revealed either high expression of REST4,

lack of NRSF expression, or lack of expression of the NRSF cofactor SWI/SNF

complex, all potentially leading to reduced NRSF activity in these cancer cells.56

In

human mammary epithelial cells, an RNAi-based screen identified NRSF as one of

five genes that lead to transformation in the form of anchorage-independent

proliferation when expression was lost. Colon cancer tissues and cell lines were

searched for chromosomal aberrations that would lead to loss of heterozygosity of the

identified potential tumor suppressors, leading to identification of 34 genes, including

NRSF, commonly mutated in colon cancer. Deletions affecting NRSF or a frameshift

mutation leading to a truncated version of NRSF were found in a significant portion of

the colon cancer primary samples and cell lines studied. Ectopic NRSF expression in

a colon cancer cell line that had lost NRSF significantly reduced proliferation, while

expression of the truncated NRSF mutant protein in mammary epithelial cells led to

transformation, suggesting that the truncated protein functions as a dominant negative.

25

Loss of NRSF expression was found to increase the anchorage-independent growth of

mammary epithelial cells through stimulation of the phosphoinositide 3-kinase

pathway.57

Although NRSF is mutated or downregulated in several cancers, it was

actually found to be highly expressed and binding to NRSEs in medulloblastoma cells.

An assay using an NRSE-containing reporter showed strong NRSF-mediated reporter

repression in medulloblastoma cells when compared to neuronal progenitors or

differentiated neurons. Expression of an NRSF-VP16 fusion or an NRSF dominant

negative in medulloblastoma cells and established tumors through transient

transfection or adenoviral infection resulted in increased expression of NRSE-

containing genes and neuronal differentiation genes, eventually halting tumor growth

and causing massive apoptosis.58

NRSF alone is not sufficient to cause tumorigenesis

as constitutive NRSF expression in neuronal cells or neurons of transgenic mice did

not lead to tumors. Coordinated overexpression of Myc and NRSF, along with an

appropriate local environment, is likely needed in order for tumors to develop.

Cellular context is critical to NRSF function as either a tumor suppressor or an

oncogene. In cells that normally have NRSF expression and repression of neuronal

genes, such as epithelial cells, NRSF functions as a tumor suppressor and its loss leads

to abnormal expression of some neuronal genes, causing the cells to more closely

resemble neural progenitors and continue to divide. In differentiating or terminal

neuronal cells, NRSF is not normally expressed or does not function as a repressor. In

this context, NRSF functions as an oncogene and its abnormal expression blocks

neuronal differentiation and allows proliferation and tumorigenesis.56,59

26

NRSF has also been found to play a role in Huntington disease. This was

discovered through the observation that wildtype huntingtin protein increases

transcription of BDNF, a factor important to the survival of straital neurons that

contains an NRSE in one of its promoters. Reporter constructs containing the BDNF

NRSE were transfected into ST14A rat striatal neural cells expressing either wildtype

or mutant huntingtin and into neural cell lines derived from mice with a CAG

expansion knock-in in the huntingtin gene. The presence of wildtype huntingtin

increased reporter activity while mutant huntingtin reduced reporter activity, both in

an NRSE-dependent manner. The reporter activity was directly proportionate to the

level of wildtype huntingtin in the cell and could be rescued by expression of

exogenous wildtype huntingtin in a mutant background, indicating that it is the loss of

wildtype huntingtin rather than the presence of mutant huntingtin that causes the

reduction of BDNF transcription in Huntington disease. EMSA identified strong

NRSF binding to the BDNF NRSE from the cytoplasm of cells with wildtype

huntingtin, but only weak binding in the nuclear fraction. This pattern was reversed in

cells with mutant huntingtin, showing weak cytoplasmic binding but strong nuclear

binding. These assays indicated that huntingtin affects BDNF transcription by

recruiting NRSF to the cytoplasm, thus releasing its repression of BDNF. Mutant

huntingtin is no longer able to recruit NRSF, allowing it to remain in the nucleus

where it continues to repress BDNF transcription. Western blotting and

immunofluorescence verified this differential localization of NRSF in wildtype and

disease cells, and direct binding between wildtype huntingtin and NRSF was shown

by coIP in neuronal cells and mouse and human brain extracts. Through ChIP and

27

RT-PCR, several NRSE-containing genes were found to be bound by NRSF and

repressed in two different Huntington disease models, neuronal lines developed from

huntingtin expansion knock-in mice and cerebral cortex tissue from the knock-in mice

and from transgenic mice expressing huntingtin with a 150 glutamine expansion.

Using ES cells and mice which had either one or both huntingtin alleles inactivated,

NRSF binding at target genes was found to correspond proportionately to the

concentration of wildtype huntingtin. Expression of a NRSF dominant negative in

Huntington disease model neuronal cells was able to rescue expression of several

NRSF target genes. This same pattern of increased NRSF binding at target genes in

neuronal tissues was seen in human Huntington disease samples, and ChIP on chip

identified this increased binding at genes encoding ion channels, adhesion molecules,

and proteins involved in synaptic activity, signal transduction, metabolism, and

neutrophins. The direct association between huntingtin and NRSF may explain the

neuronal-specific phenotype of Huntington disease as disregulation of NRSF target

genes is likely to most seriously impact neurons.60,61

In the still emerging research on eukaryotic negative regulation, NRSF is a

relatively well-studied example, but questions remain. While repression of neuronal

genes in non-neuronal cells is likely to be a major function of NRSF, evidence of its

presence and activity in neuronal cells suggest additional roles. Fortunately, thorough

characterization of the protein and binding site provides strong tools for genomic

research. Consideration of NRSF binding and function across the genome in a variety

of cell types is the best way to identify and clarify the various activities of this factor.

28

Chapter 2: Comparative genomics modeling and experimental validation of

NRSEs

Citation:

Mortazavi, A., Leeper Thompson, E.C., Garcia, S.T., Myers, R.M. & Wold, B.

Comparative genomics modeling of the NRSF/REST repressor network: From single

conserved sites to genome-wide repertoire. Genome Res. 16, 1208–1221 (2006).

Although there have been several genome-wide computational searches for

NRSEs, there is relatively little experimental validation of the binding sites in cells.

To further understand NRSF function, a more complete picture of its genome-wide

targets was needed. I used chromatin immunoprecipitation followed by quantitative

PCR analysis of individual candidate binding sites, a relatively high-throughput

method, to identify as many true NRSEs as possible. In this effort, I was joined by a

collaborator, Ali Mortazavi, who would provide computationally identified candidate

sites and use the resulting experimentally validated binding sites to further refine the

program for NRSE discovery. Based on the ChIP data, the program was found to be

an effective tool for identifying true binding sites genome-wide.

29

Conservation-based genomic computational prediction of NRSEs

Our collaborator, Ali Mortazavi, developed a set of open-source software tools,

called Cistematic, to derive a binding site model based on one or more functionally

tested and conserved binding sites and to find matching sites genome-wide that are

most likely to be functional based on conservation between the mouse, human, and

dog genomes. The focus on conservation across genomes is based on the idea that a

functionally important transcription factor binding site will be more likely to be

conserved, likely along with surrounding DNA and its target gene, than a

nonfunctional site. The first step of the process to finding NRSEs genome-wide is to

derive a position specific frequency matrix (PSFM) model for the binding site of

interest. Three NRSE models were created. The first PSFM was built using orthologs

from one gene, SCG10 (STMN2), in the mouse, human, and dog genomes. This model

was then used to search for similar sites within larger conserved domains in at least

two of the three genomes. The resulting matches were used to refine the original

PSFM into a model called NRSE2. The second model, nrsePWM33, was built from a

collection of 33 known NRSEs (Figure 1). The final model was based on several

other individual NRSEs. All three models were extremely similar, indicating that the

method derives convergent binding site models from different NRSE starting sites

(Figure 2).

Determining the best candidates for functional NRSEs from an initial set of

NRSE matches requires the application of a similarity threshold for inclusion. Past

binding data for individual NRSEs was used to set an initial membership threshold to

be applied to NRSE2 matches. The data includes several validated NRSEs as well as

30

several “false positives”, sites that resemble the NRSE but have been shown not to be

bound by NRSF. When these data were plotted as a function of PSFM match score, a

threshold of 84% match score appeared to be the best initial limit. Interestingly, the

PSFM match score was also significantly correlated with repression strength as

recorded in a reporter transfection assay (Figure 3).37

Finally, NRSE2 was used to search the mouse, human, and dog genomes for

matches that fell above the 84% threshold. All genes within a 10 kb radius of each

match were grouped into regulatory cohorts. The group of human genes was further

refined by requiring that NRSE2 matches also existed within 10 kb of an ortholog in

the mouse or dog genome, resulting in a cohort of 660 genes. Conservation was not

required outside of the NRSE2 match sites. The NRSE2 human gene cohort was

compared with those found in previous studies37,38

and was found to contain all the

genes in the previous cohorts. Forty percent of NRSE2 matches were found to be

within 5 kb of transcription start sites, however 25 percent of matches are greater than

10 kb from the start or end of any gene.

31

Figure 1: Computational approach to NRSE PSFM refinement

(A) Results from genome-wide matches to the initial NRSE PSFM (SCG10) were analyzed with

cisMatcher and used to create a refined NRSE PSFM (NRSE2). (B) A refinement starting with a PSFM

of 33 known sites produces a result very similar to NRSE2.

32

Figure 2: Different seed motifs converge following motif refinement

(A) A total of 10 initial seed motifs from known or predicted sites are compared using the motif

similarity score to our starting motif (SCG10) as well as a PSFM of 33 known instances (NRSEpsfm33)

and its refined version (NRSEpsfm33+R). The correlations median is 0.80. (B) Motif refinement of

SCG10 (called NRSE2) and of the 10 initial motifs (denoted with a +R) are markedly more similar,

with a motif correlations median of 0.91.

Figure 3: Selection of a threshold for NRSE2 and correlation of score with

repression

(A) Thirty-three known instances (triangles) and four false positives (circles) were scored with the

NRSE2 PSFM using a consensus score. A threshold of 84% of the best possible score was selected

conservatively to exclude the false positives, also excluding about 6% of true positives. (B) The

consensus scores of 10 known instances and three false positives were plotted against their relative

repression in a transient transfection reporter assay.37

100% and above reporter activity represents no

repression. R2 = 0.82

33

Chromatin immunoprecipitation analysis of predicted NRSEs

Before testing the computationally derived NRSE candidates, I checked that I

would be able to replicate past NRSF binding results using a mouse monoclonal anti-

NRSF antibody in chromatin immunoprecipitation (ChIP). Using two preparations of

chromatin from two separate growths of Jurkat cells, I performed a ChIP followed by

quantitative PCR using primers designed to flank several true NRSE sites. I also

tested a site known not to be bound, but having sequence similarity to NRSE. The

PCR enrichment of each NRSE-flanking primer pair was calculated over the

enrichment using negative primer pairs. Significant enrichment over the negatives

was seen for all of the known NRSE sites tested, while the false positive site was not

significantly enriched (Figure 4). This gave me confidence that the anti-NRSF

antibody would perform accurately in the context of ChIP.

Next, I tested a large number of the NRSE2-matched candidates, and 113

potential sites were chosen for experimental validation. In order to have validations

that would help to further refine the computational matching process, I chose 42

matches that fell below the initial 84% PFSM match threshold. I made two chromatin

preparations from two growths of Jurkat cells. These preparations were large enough

to ensure that all 113 candidates could be assayed for binding in both preparations.

Primers flanking each candidate site were designed and tested by use in quantitative

PCR on a standard curve of genomic DNA. The PCR enrichment of each candidate

was calculated over the same negative primers. Enrichment was considered

significant if it was greater than three standard deviations from the mean of the

enrichments of the negative primers. Of the 71 candidates that had a PSFM match

34

score above the 84% threshold, 70 were bound by NRSF. Of the 42 candidates that

had a match score below the 84% threshold, 29 were negative for NRSF binding. The

chromatin immunoprecipitation data show that the 84% threshold is conservative and

best to ensure minimal false positives at the expense of missing some true sites (Figure

5). Depending on the focus of future studies, the threshold could be adjusted. A Gene

Ontology (GO) analysis of the 660 genes in the NRSE2 cohort showed significant

enrichment with P-values less than 1e-6 in categories such as “synaptic transmission,”

“neurogenesis,” and “transporter activity.” These functional terms support the theory

that NRSF is a repressor of neuronal genes.

Figure 4: Validation of anti-NRSF antibody for use in ChIP

Monoclonal mouse anti-NRSF antibody was used in ChIP of chromatin preparations of two different

growths of Jurkat cells. Quantitative PCR was performed using primers flanking known true NRSEs

and one false positive. The first 11 genes listed indicate the significant PCR enrichments found at their

known NRSF binding sites. The TRBC1 site, a false positive, did not show significant enrichment over

the five negative primers used.

0

10

20

30

40

50

60

70

80

90

100

GR

IA2

TU

BB

3

OR

1E

1

GA

D1

GLR

A1

NE

F3

CY

P1

…

SC

N2A

2

L1C

AM

SC

G10

ZN

F175

TR

BC

1

neg1

neg2

neg3

neg4

neg5

fold

enrichm

ent

Replicate 1

Replicate 2

35

Figure 5: Quantitative analysis of NRSF ChIP

A total of 113 candidate NRSE2 matches, 42 of which fell below the 84% threshold (green vertical

line), were assayed for NRSF binding using ChIP followed by quantitative PCR. Fold enrichment was

calculated by dividing the amount of amplified DNA in each reaction, found for each primer pair by

using a genomic DNA standard curve, by the mean of the recovered amounts of five negative primers

matching nongenic, nonconserved regions. Fold enrichments above three standard deviations from the

mean of the five negatives (red horizontal line) were considered to be bound sites. An exponential

regression (black line) has R2 = 0.56. Thirteen of the 83 bound sites fell below the 84% threshold.

Analysis of NRSE2 matches associated with microRNAs

Twenty-one microRNAs, representing 16 microRNA families, were found to

be within 25 kb of NRSE2 matches. All but one of these microRNA families had been

shown to be expressed in neurons during differentiation. Six of the microRNA

families found are categorized to be “brain specific” or “brain enriched.”62

Seven of

the microRNAs are located in introns of NRSE2 cohort genes and may be regulated as

part of the protein-coding gene. These include miR-153 in PTPRN, miR-139 in

PDE2A, miR-9-1 in CROC4, miR-7-3 in C19orf30, and miR-24-1, miR-27b, and miR-

23b in C9orf3. I tested eleven of the twenty-one microRNAs by ChIP followed by

36

quantitative PCR and found that ten were bound by NRSF in Jurkat cells (Table 1). In

lists of predicted target RNAs for microRNAs, we found that three of the microRNA

families (miR-29b, miR-124a, and miR-153) may target CoREST, while miR-153

may also target NRSF itself.63

Thus, interactions between NRSF, CoREST, NRSEs

regulating the microRNAs, and the microRNAs could result in a feedforward loop for

more efficient downregulation of NRSF once NRSF-mediated repression has begun to

decrease (Figure 6).64

Table 1: MicroRNAs with associated NRSE2 matches are expressed in brain

MicroRNAs with an NRSE2 match within 25 kb are shown along with their expression pattern in

human and mouse brain, and mouse P19 and human NT2 cell lines during retinoic acid-induced

neuronal differentiation. MicroRNAs in bold are categorized as “brain specific” or “brain enriched.”62

Groups of microRNAs near a single NRSE are designated by the same “clust” label. Asterisks indicate

members of the same microRNA family that have only one member tested for expression pattern, and

are all listed with the same pattern. NRSEs with ChIP enrichments higher than 2.44 are considered

bound.

37

Figure 6: NRSF gene regulatory model

(A) NRSF in conjunction with CoREST and other co-repressors prevents the transcription of several

hundred targets, including neuronal splicing factors, transcription factors, and microRNAs, as well as

many terminal differentiation genes in the stem cell. (B) Upon receiving neurogenic signals to

terminally differentiate, the NRSF protein is degraded, leading to derepression of its targets, which are

now available to activators. In particular, the NRSE-associated miR-153, located within the pan-

neuronal gene PTPRN that has an NRSE in one intron, is predicted to down-regulate NRSF and

CoREST mRNAs, thus maintaining derepression.

38

Chapter 3: Comparison of genome-wide NRSF binding in neuron-derived and

non-neuronal cells

Although NRSF was initially identified as a repressor of neuronal genes in

non-neuronal cells, studies have found NRSF expression and NRSF-mediated

regulation of several individual genes in neuronal cells. These findings indicate that

repression of neuronal genes is not the only function of NRSF and that NRSF targets

or activity must differ between neuronal and non-neuronal cells. I performed ChIP

and ultrahigh-throughput sequencing on non-neuronal and neuron-derived human cell

lines to identify the differences in NRSF binding pattern. Determined binding sites

were associated with nearby genes. I found that a majority of the NRSF-occupied

sites in the neuron-derived cell lines were also occupied in non-neuronal lines, but

over half of the sites occupied in non-neuronal cells were unoccupied in the neuron-

derived lines. Gene Ontology analysis of genes associated with occupied sites in all

but one cell line, a neuron-derived line, showed significant overrepresentation of

neuronal terms. However, genes associated with sites unique to the neuron-derived

lines showed overrepresentation of terms involved in transcriptional and translational

regulation. Motif analysis identified only the canonical NRSE in binding sites

common to all of the cell lines and a lack of any recognizable motif in binding sites

unique to each cell line. This analysis demonstrates that there is a group of binding

sites that contain the canonical NRSE and is bound by NRSF regardless of cell type.

Binding sites unique to each cell line are bound by NRSF in the absence of the NRSE

and are associated with genes that are more likely important to cell type specific

functions rather than determination of neuronal identity.

39

The monoclonal mouse anti-NRSF antibody performed exceptionally well in

ChIP and allowed me to identify a large number of true binding sites in Jurkat cells.

With this reliable tool in hand, I investigated NRSF binding sites in neuron-derived

cells. This was of particular interest because a complex role for NRSF in neurons had

been hinted at in previous studies, but all genomic binding site analyses had focused

on non-neuronal cells. At this time, ultrahigh-throughput DNA sequencing

technology became available, allowing a direct assay of every binding site in the

genome. With no reliance on predictions of binding sites and the ability to sequence

millions of pieces of DNA at once, a much more complete and accurate picture of

NRSF targets is possible. The Myers lab developed an assay using ChIP followed by

ultrahigh-throughput sequencing, called ChIP-seq, and validated the method using the

monoclonal mouse anti-NRSF antibody on Jurkat cells. When the peaks found by the

new method were compared to NRSF binding sites previously found by ChIP

followed by quantitative real-time PCR or by transfection assays, ChIP-seq was found

to have a sensitivity of 87% and a specificity of 98%. Of 754 ChIP-seq peaks

compared to computationally determined NRSE location, 94% were found to be

within 50 base pairs of the center of an NRSE sequence. Almost all NRSEs

previously found to match an NRSE PSFM at 90% or more were detected as occupied

sites in ChIP-seq. This validation study found that ChIP-seq measurements are

accurate and statistically robust, provide a high resolution localization of the binding

site, and are genome-comprehensive and sampled deeply enough to identify most sites

found by other criteria.65

40

ChIP-seq analysis of human neuron-derived and non-neuronal cell lines

To capture some variation that may occur between cell lines, I used two

neuron-derived cell lines in my analysis, the neuroblastoma lines BE(2)-C and HTB-

11. I also included the PFSK-1 neurectodermal tumor line and the U-87 glioblastoma

line as examples of non-neuronal brain cells. Finally, I used the PANC-1 pancreatic

cell line as a non-neuronal cell line. Although HTB-11 and BE(2)-C cells are not truly

terminally differentiated neurons, they are thought to represent cells arrested along the

path of neuronal differentiation. HTB-11 cells exhibit neuronal phenotype and

express multiple neurochemical markers. HTB-11 cells express HASH1 and

NEUROD1, genes expressed in the developing autonomic nervous system. These

cells also contain the neuron-specific action potential sodium ionophore and

occasionally develop long, delicate cell processes resembling axons.66-69

Both HTB-

11 and BE(2)-C cells have biochemical properties characteristic of neuronal cells.

They have high activity of dopamine β-hydroxylase, an enzyme found only in

sympathetic nervous tissue. Both cell lines are also able to convert tyrosine to

dopamine, choline to acetylcholine, and glutamate to γ-aminobutyric acid. In terms of

neuronal enzyme expression, HTB-11 and BE(2)-C cells have traits of both

cholinergic and adrenergic neurons.70

PFSK-1 cells should not be considered neuronal

because they do not express antigens typically found in either differentiated neurons or

glia. It is likely that these cells are representative of neuroepithelial stem cells prior to

commitment to neuronal or glial lineage.71

I generated chromatin preparations from two independent cultures of each cell

line, and used mouse monoclonal anti-NRSF antibody to perform ChIP on each

41

preparation. The resulting immunoprecipitated DNA fragments were then used to

create a library appropriate for ultrahigh-throughput sequencing using the Illumina

Genome Analyzer platform. This involves the addition of commercially prepared

adapters to the ends of the fragments, size selection, and PCR amplification of

correctly altered and sized fragments. The sequenced library fragments were aligned

to the UC Santa Cruz hg18 human reference assembly.

Library comparison and peak calling highlight similarities and differences

between cell lines

Once each library was sequenced to an acceptable depth, I used the Compare

Libraries tool available through the HudsonAlpha High Throughput Sequencing

(HTS) website72

to compare the libraries, revealing the degree of similarity or

dissimilarity between the two biological replicates of each cell line and between the

different cell lines (Table 2). This tool allows a comparison between two libraries

across the entire genome rather than just a comparison of the enriched regions, or

peaks, found in each library. Because it includes both enriched and non-enriched

areas of the libraries, this tool more accurately reflects the correlation between two

libraries than a simple survey of overlapping peaks would provide. As hoped, the

biological duplicate libraries showed strong similarity, ranging from 82% to 97%.

Interestingly, the two cell lines that would be expected to differ most from all the

others, U-87 and PANC-1, were not as dramatically dissimilar as might be expected.

U-87 ranged from 63% to 96% similarity with other cell lines, while PANC-1 ranged

from 44% to 88% similarity. In fact, the cell line that appeared least like the others

was BE(2)-C, ranging from 32% to 85% similarity. However, even the difference in

42

BE(2)-C was not dramatic in the majority of comparisons. Based on the library

comparison numbers, most of the peaks and troughs of the libraries are correlated in

magnitude and location, indicating that the distribution of sequence tags across the

genome is similar in almost all of the chosen cell lines. This suggests that NRSF

binding patterns show a great deal of similarity in the cell lines and that there may be a

large proportion of common NRSF binding sites.

The next step in the analysis was to use a peak calling program to identify the

genomic locations where the reads from each library are concentrated. A high number

of peaks in one location indicate an enrichment of the fragment of DNA mapping to

that location in the ChIP-seq library and thus, a likely NRSF binding site. I used the

Model-based Analysis of ChIP-seq data, or MACS, peak caller.73

This program uses

the bimodal distribution of sequence tags around the true binding site, seen in ChIP-

seq data, to calculate and apply a shift to the location of the tags, providing a truer

location for the binding site. MACS also uses control libraries to find and correct for

biases in the libraries and find peaks in the ChIP-seq library. Control libraries were

made from chromatin preps from each cell line that were treated in exactly the same

way as the ChIP-seq libraries, but omitting the IP step. Overlapping peaks are merged

by the MACS program and given a new peak center based on the location of most

overlap of the tags making up the merged peaks.73

43

Table 2: ChIP-seq library comparisons

HTB11

Rep1

U87

Rep1

PANC1

Rep1

BE2C

Rep1

PFSK1

Rep1

HTB11

Rep2

U87

Rep2

PANC1

Rep2

BE2C

Rep2

PFSK1

Rep2

HTB11 Rep1 -

U87 Rep1 0.96 -

PANC1 Rep1 0.86 0.81 -

BE2C Rep1 0.83 0.85 0.60 -

PFSK1 Rep1 0.92 0.92 0.85 0.76 -

HTB11 Rep2 0.95 0.91 0.85 0.77 0.96 -

U87 Rep2 0.94 0.96 0.88 0.75 0.95 0.93 -

PANC1 Rep2 0.77 0.71 0.96 0.48 0.77 0.76 0.80 -

BE2C Rep2 0.75 0.76 0.54 0.97 0.67 0.70 0.67 0.44 -

PFSK1 Rep2 0.69 0.63 0.82 0.36 0.82 0.78 0.79 0.81 0.32 -

Library comparisons take into account the similarity of sequence tag distribution across the genome.

Gray cells indicate comparisons between libraries made from biological replicates of the same cell line.

44

Table 3: Numbers of ChIP-seq reads and MACS peaks for all libraries

Aligned Reads Peaks

HTB11 Rep1 15.45M 2507

U87 Rep1 13.28M 5926

PANC1 Rep1 11.34M 4325

BE2C Rep1 13.71M 1876

PFSK1 Rep1 13.83M 8206

HTB11 Rep2 13.51M 14034

U87 Rep2 13.38M 3490

PANC1 Rep2 7.98M 3518

BE2C Rep2 10.83M 5044

PFSK1 Rep2 12.32M 6237

HTB11 TC 18.81M

U87 TC 17.01M

PANC1 TC 17.11M

BE2C TC 12.10M

PFSK1 TC 12.21M

Libraries designated “TC” are control libraries made from reverse crosslinking prepared total chromatin

(i.e. chromatin not subjected to immunoprecipitation) from each cell line and proceeding with the

library building protocol. These libraries are compared with the ChIP-seq libraries in order to find

peaks.

45

Assembly of gene cohorts associated with ChIP-seq peaks

Before searching for genes near the peaks found in the ChIP libraries, I

identified the common peaks between biological replicates of each cell line and

discarded any peaks found in only one replicate. In some cases the number of peaks

varied greatly between replicates, so I took the intersecting peaks as the best

representation of the highest confidence NRSF binding sites in each cell line. For

each peak, I searched for the closest RefSeq genes. Depending on cell line, between

50.9% and 62.8% of peaks overlapped or were within a gene. In the cases of peaks

not overlapping genes, I found the nearest upstream and downstream genes. Next, for

each paired ChIP-seq peak and gene, I found the distance between the center of the

peak and the transcription start site of the gene. Any peak that was not within 10 kb of

a transcription start site was filtered out. The resulting peaks had an average distance

of between 1 kb and 2.5 kb from the transcription start site of the associated gene.

I next focused on the differences between the gene cohorts found in the

neuron-derived cell lines and those found in non-neuronal lines PANC-1, PFSK-1, and

U-87. 65% of the combined gene cohort of the three non-neuronal lines was unique

from the combined gene cohort of BE(2)-C and HTB-11, while only 31.5% of the

neuron-derived cohort was unique from the non-neuronal cohort. This suggests that a

majority of the genes with a nearby NRSF-occupied NRSE site in non-neuronal cells

are not bound by NRSF in neuron-derived cell lines, but that most of the genes

associated with an occupied NRSE in neuron-derived cell lines are similarly bound in

non-neuronal lines. I also identified gene cohorts unique to each individual cell line

and the gene cohort commonly bound in all of the cell lines. The 224 genes in the

46

cohort common to all the cell lines were associated with ChIP-seq peaks that had an

average score of 2063.61, while the unique gene cohorts had associated peaks with an

average score ranging from 192.49 to 345.54. This shows that commonly bound

genes are associated with very high scoring peaks while uniquely bound genes are

associated with much lower scoring peaks. This may mean that the strongest NRSF

binding sites are occupied regardless of cell type.74

I subjected the gene cohort from each cell line, the genes unique to each cell

line, the gene cohort common to all of the cell lines, the genes unique to the neuron-

derived lines, the genes unique to the non-neuronal lines, the genes unique to and

common to all neuron-derived lines, and the genes unique to and common to all non-

neuronal lines to Gene Ontology (GO) analysis to see if greatly differing gene

functions would be highlighted. Each cohort was ordered in descending order of

associated peak height, so that genes associated with the most commonly bound or

most strongly bound sites were at the top. The analysis took this order into account.

Expected neuron-related terms such as “synapse,” “channel activity,” and

“transmission of nerve impulse” were among the top ten terms in the PANC-1 and

HTB-11 cohort analyses, and various terms relating to transporters and channel

activity appeared in the top ten terms for all of the cell lines except for BE(2)-C.

Terms related to translation and the ribosome figured prominently in the BE(2)-C,

BE(2)-C unique, PFSK-1 unique, and neuron-derived unique cohorts. While the

cohort of genes common to all of the cell lines was also connected to some neuronal

terms such as “cell-cell signaling” and “channel activity”, the cohorts unique to each

cell line had a wide spread of terms which may be related to cell type-specific

47

functions. Interestingly, the GO terms most overrepresented in the HTB-11 unique

gene cohort related to transcription factor activity and chromatin configuration (Table

4).75

This analysis suggests that the commonly bound and most strongly bound NRSF

target genes are involved in neuronal functions while NRSF targets specific to neuron-

derived cell lines have a roles in transcriptional regulation, chromatin remodeling, and

regulation of translation and translational machinery.

Table 4: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts

BE2C HTB11 PANC1 PFSK1 U87

1 ribosome ion channel activity ion channel activity ion channel activity gated channel activity

2 cytosolic ribosome

substrate specific channel activity

substrate specific channel activity

substrate specific channel activity

ion channel activity

3 cytoplasmic part channel activity channel activity channel activity substrate specific channel activity

4 ribonucleoprotein complex

passive transmembrane transporter activity

passive transmembrane transporter activity

passive transmembrane transporter activity

ion transport

5 organelle

gated channel activity

gated channel activity

gated channel activity

channel activity

6 structural constituent of ribosome

ion transport multicellular organismal process

cation channel activity

passive transmembrane transporter activity

7 intracellular organelle synapse

cation channel activity

ion transport cation channel activity

8 intracellular part

cation channel activity synapse

metal ion transmembrane transporter activity

metal ion transmembrane transporter activity

9 cytoplasm

ion transmembrane transporter activity

ion transport ion transmembrane transporter activity

ion transmembrane transporter activity

10 translation

transmission of nerve impulse

transmission of nerve impulse

metal ion transport transmembrane transporter activity

48

Table 4: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts

(cont’d)

Common to All Cell Lines

BE2C Unique HTB11 Unique

PANC1 Unique

PFSK1 Unique

U87 Unique

1 localization guanyl nucleotide binding

sequence-specific DNA binding

thiamin diphosphokinase activity

intracellular part

protein phosphatase regulator activity

2 membrane part GTP binding nucleosome

thiamin diphosphate biosynthetic process

intracellular organelle

phosphatase regulator activity

3 cell-cell signaling

guanyl ribonucleotide binding

nucleic acid binding

thiamin diphosphate metabolic process

organelle protein phosphatase type 2A complex

4 ion channel activity

cellular process

transcription regulator activity

thiamin & derivative biosynthetic process

protein binding

protein phosphatase type 2A regulator activity

5 establishment of localization

macromolecular complex

DNA binding multicellular organismal development

cytoplasm

protein serine/threonine phosphatase complex

6 intrinsic to membrane

cellular biosynthetic process

nucleosome assembly

thiamin & derivative metabolic process

intracellular membrane-bounded organelle

selenide, water dikinase activity

7 substrate specific channel activity

mitochondrion chromatin assembly

diphospho transferase activity

intracellular

phospho transferase activity, paired acceptors

8 integral to membrane

ribosome

transcription factor activity

multicellular organismal process

membrane-bounded organelle

enzyme regulator activity

9 channel activity organelle

DNA packaging

developmental process

Macro molecule metabolic process

cholinesterase activity

10 passive transmembrane transporter activity

legumain activity

chromatin assembly or disassembly

cell-cell adhesion

ribosomal subunit

49

Table 4: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts

(cont’d)

Neuron-derived Unique

Non-Neuronal Unique

Neuron-derived

Common & Unique

Non-Neuronal Common &

Unique

1 organelle protein binding nucleic acid binding

helicase activity

2 intracellular organelle

intracellular part Nucleus ATP-dependent helicase activity

3 metabolic process intracellular organelle DNA binding

tRNA-pseudouridine synthase activity

4 intracellular membrane-bounded organelle

organelle intracellular membrane-bounded organelle

pseudouridine synthase activity

5 membrane-bounded organelle

developmental process

membrane-bounded organelle

anatomical structure development

6 intracellular part

multicellular organismal development

intracellular organelle

7 cellular metabolic process

cytoplasm organelle

8 gene expression

intracellular membrane-bounded organelle

protein-DNA complex assembly

9 ribosome

membrane-bounded organelle

regulation of cellular process

10 intracellular organelle part

macromolecule metabolic process

regulation of cellular metabolic process

50

Motif analysis of ChIP-seq peaks in each cell line

Finally, I used MEME software to analyze the sequences of the peaks found to

be within 10 kb of a gene in each of the cell lines. Regardless of cell line, the only

motif identified was the canonical NRSE. This was also true when I analyzed the

sequences of peaks common to all of the cell lines (Figure 7). However, when I

analyzed the peaks unique to each cell line, peaks unique to neuron-derived lines,

peaks unique to non-neuronal lines, or peaks unique to and common to all neuron-

derived lines, the canonical motif was not found and the software primarily returned

highly repetitive mononucleotide and dinucleotide sequences. Interestingly, the

canonical NRSE was identified as the only motif present in the peaks unique to and

common to all non-neuronal lines, and a motif resembling an NRSE half-site was

found only in the peaks unique to PANC-1 (Figure 7). As common peaks are also

highly scoring peaks and unique peaks are also lowly scoring peaks, it is possible that

the canonical NRSE results in stronger NRSF binding, while weaker NRSF binding is

possible at some locations even in the absence of a canonical NRSE motif or any

recognizable motif.

51

Figure 7: MEME motif analysis of NRSF peak sequences

The first motif is the canonical NRSE. It was identified as a significant motif when all peak sequences

from any cell line were included in the MEME analysis. It was also identified as a significant motif in

the analysis of peaks common to all of the cell lines and peaks unique to and common to all the non-

neuronal cell lines. The second motif is a NRSE half site identified as a significant motif among the

peaks unique to PANC-1.

52

Chapter 4: Expression analysis of potential NRSF target genes in neuron-derived

and non-neuronal cells

Although some NRSF-occupied sites in neuron-derived cell lines were found

to be specific to those cell lines, a majority of the occupied sites were found to also be

occupied in non-neuronal cell lines. The genes associated with this group of

commonly bound sites include and overrepresentation of genes important to neuronal

function. This finding implies that NRSF may not be acting as a repressor at all

targets in all cell lines. Past studies have revealed NRSF functioning as an activator in

neuronal cells. I used RNA-seq to determine the expression level of genes associated

with NRSF binding sites in each cell line. I found that there were a small number of

associated genes in each cell line that were highly expressed, although the magnitude

of this expression varied greatly between cell lines. Genes associated with sites

commonly bound in all cell lines were more highly expressed in the neuron-derived

lines, and genes associated with sites uniquely bound in one cell line tended to be

more highly expressed in the cell line in which they were bound. Although

experiments involving perturbation of NRSF binding would be required to establish

causation, these observations suggest that NRSF binding is not associated with strong

repression in all cases and may be associated with activation.

Once I had identified NRSF binding sites in several cell lines and constructed

the associated gene cohorts, I investigated the expression level of these potential

NRSF target genes. Because NRSF had been found to function as both a repressor

and an activator in different genetic and cellular contexts, I asked if all of the potential

target genes I had identified had low expression when the nearby NRSE was bound by

53

NRSF or if some were highly expressed in the bound state, providing support for a

possible activating function of NRSF. There was no experimental perturbation of

NRSF expression or function during the studies I report here, so the direct effect of

NRSF binding on target gene expression in each cell line cannot be deduced.

However, the expression of a gene that has a nearby occupied NRSE in one cell line

can be compared with its expression in another cell line in which the same NRSE is

unoccupied, and this may provide some information about the effect of NRSF binding

on the expression of the gene. To achieve this, I used ultrahigh-throughput DNA

sequencing technology, utilizing a protocol that would provide quantitative measures

of all mRNA transcripts in each cell line. This protocol, called RNA-seq, was

validated using adult mouse brain, liver, and skeletal muscle tissues and identification

of known, in vitro transcribed RNA standards introduced into the samples in amounts

spanning a wide range of abundance. RNA from a well-characterized muscle-specific

gene was easily found in the muscle sample, but absent from the other tissue samples.

93% of all the RNA-seq reads mapped to known and predicted exons, while only 3%

mapped to intergenic regions. The RNA-seq data for the added RNA standards was

linear across a range of five orders of magnitude of RNA concentration. Sequence

coverage was highly reproducible and uniform, and transcript detection was robust

even at a concentration calculated to correspond to only one transcript per cell.76

RNA-seq analysis of human neuron-derived and non-neuronal cell lines

I created RNA-seq libraries from two independent cultures of each cell line, as

I did with the ChIP-seq libraries. Two consecutive selections using Oligo(dT)

magnetic beads separated the mRNA transcripts from the total RNA that had been

54

extracted from each cell growth. These transcripts were then fragmented through

controlled hydrolysis, a technique found to improve uniformity of sequence coverage

and alleviate overrepresentation of 5’ ends and favored random priming sites. These

fragmented transcripts were then used to synthesize cDNA using random hexamer

primers. The resulting cDNAs were processed into a library appropriate for ultrahigh-

throughput sequencing using the Illumina Genome Analyzer platform by the same

protocol used to process the ChIP fragments for ChIP-seq analysis. Once the

fragments were sequenced and aligned to the genome, the sequence tags were further

analyzed by Enhanced Read Analysis of Gene Expression (ERANGE) software. This

program assigns reads to their unique site of origin in the genome or to their most

likely site of origin in the case of reads matching to multiple locations, detects reads

that span splices and assigns them to their gene of origin, collects reads that cluster

together but do not map to a known exon into candidate exons, and calculates the

prevalence of transcripts assigned to known or candidate RNAs. Finally, the

expression level of each transcript is calculated and expressed as a function of both

molar concentration and transcript length. This is quantified in reads per kilobase of

exon model per million mapped total reads in the sample, called RPKM (Figure 8).76

55

Figure 8: Calculation of RPKM

The calculation of RPKM takes into account the length of each exon model and the total number of

mapped reads in the sample.

Correlation of peak strength and expression level of associated genes in each cell

line

Once I had attached the RPKM found in RNA-seq to each of the genes

associated with an NRSF ChIP-seq peak, I wanted to see if these two measurements

were correlated in any way. I graphed ChIP-seq peak strength against expression

level, expressed as RPKM, for each gene in the complete cohorts for each cell line. I

found that peak strength is not correlated to expression level of the nearby gene.

These graphs highlight genes that have high expression even when NRSF is bound

nearby. The majority of cohort genes in all cell lines have very low expression as

would be expected if NRSF functions as a powerful repressor. Although the gene

cohorts in each cell line had a mean RPKM between 12.7 and 55.2, several genes had

56

an RPKM of 1000 or greater in BE(2)-C, HTB-11, and PFSK-1 (Figure 9). This

highly expressed subgroup included CHGA, RPLP1, and EEF2, in BE(2)-C, CHGA,

DBH, RPL38, RPS4X, and EEF2 in HTB-11, and VGF, RPS6, RPLP1, EEF2,

GNB2L1, RPL36, SPP1, ENO1, RPL30, ETV4, PFN1, ACTG1, and RPS5 in PFSK-

1. While genes with the highest expression in the BE(2)-C and HTB-11 cohorts

ranged up to an RPKM of 3502.64 and 3233.93 respectively, the cohort gene with the

highest expression in PFSK-1, VGF, had an RPKM of 5638.23. Interestingly, the

genes with the highest expression in the PANC-1 and U-87 cohorts have an RPKM of

only 463.95 and 885.03 respectively. This demonstrates that some genes can be

expressed at a high level even when NRSF is bound nearby, but that the magnitude of

this high expression varies widely by cell line.

57

Figure 9: Correlation between peak strength and expression of associated gene

R² = 0.0093

0

500

1000

1500

2000

2500

3000

3500

4000

0 500 1000 1500 2000 2500 3000 3500

Exp

ress

ion

(R

PK

M)

Peak Strength

BE(2)-C Binding & Expression

R² = 0.0005

0

500

1000

1500

2000

2500

3000

3500

0 500 1000 1500 2000 2500 3000 3500

Exp

ress

ion

(R

PK

M)

Peak Strength

HTB-11 Binding & Expression

58

Figure 9: Correlation between peak strength and expression of associated gene

(cont’d)

R² = 0.0118

0

50

100

150

200

250

300

350

400

450

500

0 500 1000 1500 2000 2500 3000 3500

Exp

ress

ion

(R

PK

M)

Peak Strength

PANC-1 Binding & Expression

R² = 0.00490

1000

2000

3000

4000

5000

6000

0 500 1000 1500 2000 2500 3000 3500

Exp

ress

ion

(R

PK

M)

Peak Strength

PFSK-1 Binding & Expression

59

Figure 9: Correlation between peak strength and expression of associated gene

(cont’d)

Expression of genes in cell line unique, common, neuron-derived, and non-

neuronal cohorts across all cell lines

Comparison of the expression of genes uniquely bound in each cell line with

the expression of those same genes in the other cell lines in which they are not bound

by NRSF provides clues about the possible effects of NRSF binding on the expression

level of a neighboring gene. I created heatmaps that included the expression across all

five cell lines of the genes uniquely bound by NRSF in each cell line. Based on these

heatmaps, which included only the gene cohorts unique to each cell line and thus the

genes associated with the generally weaker NRSF ChIP-seq peaks, it appeared that

NRSF binding near a gene was associated with a tendency for the gene to be expressed

at a higher level than it was in cell lines with a lack of NRSF binding near the same

gene (Figure 10). However, this was not an overwhelmingly obvious trend. I also

R² = 0.0064

0

100

200

300

400

500

600

700

800

900

1000

0 500 1000 1500 2000 2500 3000 3500

Exp

ress

ion

(R

PK

M)

Peak Strength

U-87 Binding & Expression

60

created a heatmap of the expression of all genes commonly bound by NRSF in all of

the cell lines. Most of these genes seemed to be more highly expressed in BE(2)-C

and HTB-11 than in the other three cell lines regardless of the fact that they were

NRSF-bound in all of the cell lines. The same type of heatmap was created for the

gene cohort unique to and common to the neuron-derived lines and for the gene cohort

unique to and common to the non-neuronal lines. While the neuron-derived unique

and common genes showed a trend of being more highly expressed in the bound state,

the non-neuronal unique and common genes showed no clear trend of differential

expression in response to differential NRSF binding (Figure 10). Because the

expression of these genes is being compared across different cell lines, the differential

expression that corresponds to differential NRSF binding may be affected by a number

of factors other than NRSF that differ between the cell lines. Other factors may also

contribute to the lack of differential expression levels seen in some cases with

differential NRSF binding. One fact that seems clear from the heatmaps is that the

overall expression profile of the gene cohort common to all of the cell lines, which are

associated with generally stronger NRSF ChIP-seq peaks, differs between the neuron-

derived and non-neuronal cell lines. Among these genes, binding in either BE(2)-C or

HTB-11 cells is usually associated with a higher expression level, while binding in

any of the three non-neuronal lines is usually associated with a lower expression level

(Figure 10). Finally, I used the RNA-seq data to look at the expression of NRSF itself

and the expression of some of the most common NRSF co-repressors in all of the cell

lines. The expression level of NRSF, CoREST, Sin3A, HDAC1, and HDAC2 varied

61

between cell lines. Surprisingly, the NRSF expression level was quite low in all cell

lines and undetectable in BE(2)-C (Figure 11).

62

Figure 10: Expression of commonly bound, cell line uniquely bound, neuron-

derived commonly and uniquely bound, and non-neuronal commonly and

uniquely bound genes

63

Figure 10: Expression of commonly bound, cell line uniquely bound, neuron-

derived commonly and uniquely bound, and non-neuronal commonly and

uniquely bound genes (cont’d)

Blue highlighting indicates the cell line(s) in which the binding site associated with the gene is occupied

by NRSF. Red indicates higher expression and green indicates lower expression.

64

Figure 11: Expression of NRSF and co-repressors in all cell lines

Analysis of NRSF binding and target gene expression incorporating ChIP-seq

and RNA-seq data from other non-neuronal cell lines

A second analysis of the ChIP-seq and RNA-seq data from the aforementioned

cell lines was performed with the addition of data from three cell lines prepared by a

colleague in another lab using identical protocols. These non-neuronal cell lines

included HepG2 human liver carcinoma cells, GM12878 human lymphoblast cells,

and K562 human erythroleukemia cells. Data from the BE(2)-C line was excluded

from this second analysis due to the lack of NRSF expression found in that cell line,

leaving HTB-11 as the single neuron-derived cell line. In this analysis, in order to

make my data comparable to the data from the new cell lines, sequences from both

ChIP-seq replicates in each cell line were combined and peaks were called from this

combined data pool. Peaks were then associated the nearest gene within a 2 kb radius

and designated with a location with respect to the associated gene. Peaks within genes

0

5

10

15

20

25

30

NRSF CoREST Sin3A HDAC1 HDAC2

RP

KM

BE(2)-C 1

BE(2)-C 2

HTB-11 1

HTB-11 2

PANC-1 1

PANC-1 2

PFSK-1 1

PFSK-1 2

U-87 1

U-87 2

65

were designated as being within exons or introns. A peak within 2 kb of the

transcription start site was designated as being in the promoter, while a peak within 2

kb of the gene terminus was designated as being in the 3’ proximal region. Any peak

that was not within 2 kb of a gene was not associated with a gene and was designated

as being intergenic. I focused on peaks and genes that were commonly found in all

cell lines, uniquely found in each cell line, and uniquely found in neuron-derived or

non-neuronal cell lines.

Many observations related to the peaks and gene cohort common to all of the

cell lines in this second analysis agreed with observations from the first analysis. The

locations of the peaks were found to be primarily intronic and intergenic. The

heatmap of expression of the common gene cohort once again revealed that genes with

proximal NRSF binding were likely to be more highly expressed in HTB-11 than in

the other non-neuronal cell lines. However, the heatmap also highlighted a small

group of genes that had low expression in all of the cell lines. The most

overrepresented GO terms associated with the entire common gene cohort again

included terms indicative of neuronal gene function such as “synapse”, “channel

activity”, and “transmembrane transporter activity”. A GO analysis of the subset of

genes that is lowly expressed in all cell lines included many neuronal terms, but also

included terms related to sugar and carbohydrate binding and adhesion (Figure 12).

As in the previous analysis, the only motif discovered in the peaks common to all cell

lines was the canonical NRSE.

Next I examined the peaks and gene cohorts unique to each cell line. The

location distributions of the unique peaks in each cell line except HTB-11 resemble

66

the location distribution of the peaks common to all cell lines, primarily intronic and

intergenic. However, the peaks unique to HTB-11 have a completely different

distribution, primarily exonic and located in promoter regions (Figure 13). Similarly

to the first analysis, the heatmaps displaying the expression of cell line unique genes in

all of the cell lines showed that the genes were slightly more likely to be more highly

expressed in the cell line in which they were bound (Figure 14). Again, this was not a

strong trend and was not detectable for K562. While some neuronal GO terms did

appear among the overrepresented terms for the gene cohorts unique to PANC-1,

PFSK-1, and K562, most of the GO terms were connected to either more general cell

functions or cell type specific functions. For example, HTB-11 unique genes were

overrepresented for general terms such as “organelle” and “cytoplasm” while HepG2

unique genes were overrepresented for terms related to lipid metabolism and

processing (Table 5). Motif analysis of the cell line unique peaks only returned highly

repetitive sequences.

Finally, I compared the peaks and genes unique to neuron-derived cells, in this

case unique to HTB-11, to the peaks and genes unique to and common to all of the

non-neuronal cell lines. Unsurprisingly, the location distribution of non-neuronal

unique and common peaks was very similar to that of the peaks common to all of the

cell lines (Figure 15). The heatmap of expression of genes unique to and common to

non-neuronal cell lines did not reveal a clear difference in expression between the

NRSF bound and unbound states of the genes (Figure 16). GO analysis showed

enrichment for neuronal terms among the genes unique to the non-neuronal cell lines

as expected. The smaller group of 80 genes both unique to and common to all of the

67

non-neuronal lines had fewer significantly overrepresented GO terms which included

“pseudouridine synthase activity” and “steroid binding”. The lower number of

significant terms is likely due to the low number of genes in this cohort (Table 6).

68

Figure 12: Location distribution of common binding sites, expression and GO

analysis of common gene cohort

Genes common to all cell lines Genes with low expression in all cell lines

1 gated channel activity sugar binding

2 ion channel activity carbohydrate binding

3 substrate specific channel activity solute:sodium symporter activity

4 channel activity membrane

5 passive transmembrane transporter activity cell adhesion

6 synapse biological adhesion

7 ion transmembrane transporter activity neurotransmitter:sodium symporter activity

8 substrate-specific transmembrane

transporter activity

neurotransmitter transporter activity

9 ion transport plasma membrane part

10 cation channel activity

Pie chart displays the distribution of location of common peaks and P-value indicates the difference

between this distribution and the pattern that would be created by randomly placed sites. Blue

highlighting indicates the block of genes with low expression in all cell lines.

69

Figure 13: Location distribution of peaks unique to each cell line

70

Figure 14: Expression of genes associated with peaks unique to each cell line

71

Figure 14: Expression of genes associated with peaks unique to each cell line

(cont’d)

72

Table 5: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts

HTB11 Unique

PANC1 Unique

PFSK1 Unique

U87 Unique

GM12878 Unique

HepG2 Unique

K562 Unique

1 intracellular part

potassium ion binding

nervous system development

tubulin-tyrosine ligase activity

thiamin diP activity

lipid metabolic process

cellular process

2 intracellular ion transport multicellular organismal development

Rho protein signal transduction

thiamin diP biosynthetic process

sterol biosynthesis

double-stranded RNA binding

3 intracellular organelle

system development

protein binding

diacylglycerol kinase activity

negative regulation of glycoprotein biosynthesis

isoprenoid metabolic process

glutamate receptor activity

4 organelle

potassium ion transport

system development

ErbB-3 class receptor binding

thiamin diP metabolic process

steroid biosynthesis

adenosine deaminase activity

5 intracellular membrane-bounded organelle

sugar binding anatomical structure development

activation of protein kinase C activity

thiamin & derivative biosynthetic process

cellular lipid metabolic process

multicellular organismal process

6 membrane-bounded organelle

ion channel activity

polyamine biosynthetic process

growth factor activity

negative regulation of amyloid precursor protein biosynthesis

sterol metabolic process

P-P-bond-hydrolysis-driven protein trans membrane transporter activity

7 binding substrate specific channel activity

spermine biosynthetic process

regulation of glycoprotein biosynthetic process

lipid biosynthesis

macromolecule trans membrane transporter activity

8 cytoplasm

channel activity

spermine metabolic process

amyloid precursor protein biosynthesis

alcohol metabolic process

synapse

9 protein binding

passive trans membrane transporter activity

cellular process

regulation of amyloid precursor protein biosynthesis

cholesterol biosynthesis

protein trans membrane transporter activity

10 nucleus

nervous system development

cytosolic ribosome

selenide, water dikinase activity

monooxygenase activity

multicellular organismal development

73

Figure 15: Location distribution of peaks unique and common to neuron-derived

and non-neuronal cell lines

Figure 16: Expression of genes associated with peaks unique and common to

neuron-derived and non-neuronal cell lines

74

Table 6: Top ten GO analysis terms for ChIP-seq peak-associated gene cohorts

Genes unique to HTB-11 Genes unique to non-

neuronal

Genes unique & common

to non-neuronal

1 intracellular part gated channel activity tRNA-pseudouridine

synthase activity

2 intracellular synapse pseudouridine synthase

activity

3 intracellular organelle ion channel activity intramolecular transferase

activity

4 organelle substrate specific channel

activity

steroid hormone receptor

activity

5 intracellular membrane-

bounded organelle channel activity

ligand-dependent nuclear

receptor activity

6 membrane bounded organelle passive transmembrane

transporter activity steroid binding

7 binding ion transport cell-cell signaling

8 cytoplasm developmental process

9 protein binding transmembrane transporter

activity

10 nucleus multicellular organismal

development

75

Chapter 5: Discussion and future directions

These experiments have clearly shown that NRSF is expressed and binding to

target genes in at least the HTB-11 neuroblastoma cell line. The case of the BE(2)-C

neuroblastoma cell line remains a mystery. While RNA-seq detected no NRSF

mRNA in BE(2)-C cells, significant NRSF ChIP-seq peaks were called, although the

peaks are not as strong on average than the peaks in other cell lines. Further

investigation of the location of the BE(2)-C peaks and their associated genes in

comparison with those in other cells lines revealed that 88.4% to 94.8% of BE(2)-C

peaks correspond to peaks found in the other cell lines and 85% to 93.7% of BE(2)-C

cohort genes are also members of the gene cohorts of the other cell lines. It is possible

that NRSF is expressed at an extremely low level in BE(2)-C, but even this level of

expression is enough to result in some binding of the strongest or most likely NRSF

binding sites. Further analysis of NRSF expression in BE(2)-C cells will be required

to determine if the line is truly lacking in NRSF protein.

Based on the combined binding and expression data, it seems very likely that

there are many cases where NRSF binding results in high expression or at least does

not cause significant repression. Although the majority of genes associated with

NRSF binding had very low expression in all of the cell lines, several genes were

identified, with a range of associated peak strengths, which had RPKMs above 1000 in

three brain cell lines. EEF2, eukaryotic translation elongation factor 2, was highly

expressed in all three of these lines. RPLP1, a ribosomal protein involved in

elongation, and CHGA, a neuroendocrine secretory protein, were highly expressed in

two out of the three lines. The prominent expression of the two proteins involved in

76

elongation along with the significant translational and ribosomal GO terms found in

BE(2)-C, PFSK-1 unique, and neuron-derived unique gene cohorts of the first analysis

support a role for NRSF in regulating translational machinery. In addition to the

subset of NRSF-bound genes which are highly expressed in some cell types,

comparison of the expression of uniquely bound genes in one cell line with the

expression of those same genes in cell lines in which they are not NRSF-bound shows

a slight trend for a majority of the genes to be more highly expressed in the bound

state. Because this comparison is made between different cell lines, this increased

expression cannot be definitively attributed to NRSF binding alone as a number of

other cell line specific factors may be involved in setting the final expression level.

However, this higher expression in an NRSF-bound state paired with the range of

expression to very high levels of NRSF-bound genes within a single cell line give

support to the idea that NRSF does not function solely as a powerful repressor and

may even act as an activator in some cellular and genetic contexts.

Motif analysis of the sequences of cell line unique peaks, which are also the

weaker peaks, did not result in the discovery of a canonical NRSE or any other motif.

When all of the peaks in a cell line or the peaks common to all of the cell lines were

considered, the only motif found was the canonical NRSE. The canonical NRSE is

clearly present at a significant proportion of overall peaks and is clearly associated

with the stronger peaks. This data seems to show that NRSF can bind in the absence

of the canonical NRSE and possibly even in the absence of a recognizable motif, but

that this binding is likely to be weaker or to occur in a smaller percentage of the cells.

NRSF may have a different binding mechanism in the case of these weaker, cell line

77

specific sites, and this may also be connected to altered function. As mentioned, when

the expression of genes associated with these non-canonical, unique sites is compared

between the cell line in which they are bound and those in which they are not, the

majority tend to have a higher expression in the cell line in which they are NRSF-

bound. This may support a non-repressive function for NRSF when bound at non-

canonical sites.

Two observations indicate a dramatic difference between the neuron-derived

cell line HTB-11 and all of the other non-neuronal cell lines. The first is the location

distribution of unique peaks. The peaks unique to HTB-11 are primarily located in

exons and in the promoter, while peaks unique to all other cell lines and peaks

common to all of the cell lines are primarily located in introns and intergenic regions.

This completely different distribution with respect to associated genes may indicate

that the HTB-11 unique binding sites are in a completely different category than the

binding sites unique to non-neuronal cell lines and common binding sites, and the

position with respect to the gene could change the effect of NRSF binding on

expression. The second observation is the difference in expression levels of the genes

in the cohort common to all of the cell lines. These genes are associated with NRSF

binding in all of the cell lines, but the majority of them are more highly expressed in

HTB-11 than they are in the other non-neuronal cell lines. This is sensible given that

the overrepresented GO terms for this group of genes includes several functions

important or specific to neurons, thus repression of these genes in neuron-derived cells

could be very detrimental. A heatmap comparing the peak intensities of the common

sites in all cell lines also shows that almost all of these peaks are relatively weak in

78

HTB-11 (Figure 17). In the case of the common binding sites, HTB-11 displays

weaker or less frequent NRSF binding which may result activation or at least lack of

repression of the associated genes. This may indicate a different binding mechanism

for NRSF in HTB-11 even at the sites that it shares with non-neuronal cells. The

differential binding and effect on expression could be due to other transcription factors

binding in the same area or cofactors that bind directly to NRSF, changing its binding

affinity and effect on transcription of target genes.

Figure 17: Strength of peaks common to all cell lines

Red indicates stronger peaks and green indicates weaker peaks.

79

During the course of this study, a protocol for the differentiation of mouse ES

cells into neurons was developed that will fit perfectly into the further study of the role

of NRSF in neuronal cells. This is the first neuronal differentiation protocol to be well

suited to the use of analysis methods utilizing high-throughput sequencing because it

produces a nearly homogenous neuronal population after differentiation and could be

used to create the large number of cells required. The purity of the final neuronal

population is very important because of the sensitivity of the ChIP-seq and RNA-seq

techniques to even low numbers of chromatin fragments or transcripts, making

contaminating sources detectable unless they are at very low levels. Following this

recent protocol, mouse ES cells are cultured on feeder cells for at least two passages.

They are then cultured without feeders for at least another two passages but no more

than five passages. The cells are then forced to form free floating cellular aggregates,

or CAs, by trypsinization and plating in CA medium. This medium is changed after

two days. After four more days the media is changed and retinoic acid is added. After

an additional six days of culture, the CAs are dissociated by trysinization and pipeting

and the dissociated CAs are plated in N2 medium on plates coated with poly-DL-

ornithine hydrobromide and laminin. This causes the cells to differentiate to the

progenitor stage. The N2 medium is changed after two hours and again after one day,

followed by two days of culture. At this point, the medium is changed to complete

medium, causing the cells to differentiate into glutamatergic neurons. These neurons

can be maintained in culture for many weeks. A good quality culture was determined

to consist of 90% to 95% glutamatergic neurons as characterized by VGLUT, β-

tubulin III, MAP2, and Tau expression. The cells were found to have neuronal

80

morphology and electrical activity by about twelve days after differentiation.77

This

protocol provides an opportunity to use ChIP-seq and RNA-seq to compare NRSF

binding and effect on target gene expression before and after neuronal differentiation

within the same cell line, removing the potential problems of comparison between

different cell lines. It is possible that cells could be harvested at different points in the

differentiation, providing a way to track NRSF binding and function over the course of

differentiation.

Another future goal would be to attempt to ensure that any change in

expression of target genes is due to a change in NRSF binding status of the nearby

NRSE. Clearly an NRSF knockout in mice caused lethality so early in development

that no clear conclusions could be made. An RNAi knockdown of NRSF has been

successful in some cell lines and may be possible in the cell lines included in this

study. This technique should be attempted and would allow analysis using the ChIP-

seq and RNA-seq protocols that were used in this study. However, it is possible that

such a dramatic perturbation would not be tolerated by the cells or would have such

extensive and diverse effects as to obscure the consequence of a change in NRSF

binding status on individual genes. A more narrowly focused approach would be to

choose a group of candidate binding sites and associated genes of particular interest

and clarify the effect of loss of NRSF binding using reporter constructs. I would use

this technique to study a group of candidates chosen from the peaks common to all of

the previously studies cell lines. I would choose some binding sites associated with

lowly expressed target genes in all cell lines and other that are associated with highly

expressed target genes in HTB-11 only. This way I could study cases in which NRSF

81

may repress genes in neurons and others in which it may function as an enhancer in

neurons. These binding site sequences could then be used to build reporter constructs

containing an appropriately spaced wildtype binding site and identical constructs with

a mutated binding site. These constructs could be transiently transfected into HTB-11

or the other non-neuronal cell lines as the binding sites are bound by NRSF regardless

of cell type. Any difference in reporter expression between the wildtype and mutant

version of a construct would be due to disrupted NRSF binding.

Finally, focusing on the same candidate binding sites ranging from low to high

expression in HTB-11, I would attempt to catalog the transcription factors and

cofactors bound near or with NRSF. This could help to identify what causes the

difference in target gene expression level between the sites associated with low and

high expression. Comparing the complex of proteins bound around the NRSE

between a site in HTB-11 that results in high expression and the same site in a non-

neuronal cell line that results in low expression could also identify the cell line

specific factors that may alter NRSF function in neurons. To this end, the candidate

sequences and extracts from different cell lines could be used in a series of EMSA

supershift assays in which antibodies against candidate factors would be used to

attempt to supershift the candidate sequences in a gel shift experiment. This

experiment would begin with the known cofactors of NRSF, such as CoREST, Sin3A,

CoREST, and MeCP2 and be expanded to test for other possibilities. The most critical

future goals are to establish a direct connection between NRSF binding and a change

in target gene expression and to identify the factors in addition to NRSF that are

responsible for this change.

82

Works Cited

1. Maniatis, T., Goodbourn, S. & Fischer, J.A. Regulation of inducible and tissue-

specific gene expression. Science 236, 1237-45 (1987).

2. Quinn, J.P. Neuronal-specific gene expression - the interaction of both positive

and negative transcriptional regulators. Progress in Neurobiology 50, 363-379

(1996).

3. Johnson, P.F. & McKnight, S.L. Eukaryotic transcriptional regulatory proteins.

Annu Rev Biochem 58, 799-839 (1989).

4. Goodman, R.H. & Mandel, G. Activation and repression in the nervous system.

Current Opinion in Neurobiology 8, 413-417 (1998).

5. Ogbourne, S. & Antalis, T.M. Transcriptional control and the role of silencers in

transcriptional regulation in eukaryotes. Biochem J. 331, 1–14 (1998).

6. Johnson, A.D. The price of repression. Cell 81, 655-658 (1995).

7. Thiel, G., Lietz, M., Bach, K., Guethlein, L. & Cibelli, G. Biological activity of

mammalian transcriptional repressors. Biol. Chem 382, 891-902 (2001).

8. Schoenherr, C.J. & Anderson, D.J. Silencing is golden: negative regulation in the

control of neuronal gene transcription. Current Opinion in Neurobiology 5, 566-

571 (1995).

9. Mori, N., Schoenherr, C., Vandenbergh, D.J. & Anderson, D.J. A common

silencer element in the SCG10 and type II Na+ channel genes binds a factor

present in nonneuronal cells but not in neuronal cells. Neuron 9, 45-54 (1992).

10. Kraner, S.D., Chong, J.A., Tsay, H. & Mandel, G. Silencing the type II sodium

channel gene: A model for neural-specific gene regulation. Neuron 9, 37-44

(1992).

11. Schoenherr, C.J. & Anderson, D.J. The neuron-restrictive silencer factor (NRSF):

a coordinate repressor of multiple neuron-specific genes. Science 267, 1360-3

(1995).

12. Chong, J.A. et al. REST: A mammalian silencer protein that restricts sodium

channel gene expression to neurons. Cell 80, 949-957 (1995).

13. Scholl, T., Stevens, M.B., Mahanta, S. & Strominger, J.L. A zinc finger protein

that represses transcription of the human MHC class II gene, DPA. J. Immunol

156, 1448-1457 (1996).

14. Roopra, A., Huang, Y. & Dingledine, R. Neurological Disease: Listening to Gene

Silencers. Mol. Interv. 1, 219-228 (2001).

15. Wood, I.C., Roopra, A. & Buckley, N.J. Neural specific expression of the m4

muscarinic acetylcholine receptor gene is mediated by a RE1/NRSE-type silencing

element. J. Biol. Chem 271, 14221-14225 (1996).

16. Mieda, M., Haga, T. & Saffen, D.W. Expression of the Rat m4 Muscarinic

Acetylcholine Receptor Gene Is Regulated by the Neuron-restrictive Silencer

Element/Repressor Element 1. J. Biol. Chem. 272, 5854-5860 (1997).

17. Kallunki, P., Jenkinson, S., Edelman, G.M. & Jones, F.S. Silencer elements

modulate the expression of the gene for the neuron-glia cell adhesion molecule,

Ng-CAM. J. Biol. Chem 270, 21291-21298 (1995).

18. Bessis, A. et al. Promoter elements conferring neuron-specific expression of the

[beta]2-subunit of the neuronal nicotinic acetylcholine receptor studied in vitro

83

and in transgenic mice. Neuroscience 69, 807-819 (1995).

19. Schoch, S., Cibelli, G. & Thiel, G. Neuron-specific Gene Expression of Synapsin

I. J. Biol. Chem. 271, 3317-3323 (1996).

20. Thiel, G., Lietz, M. & Cramer, M. Biological activity and modular structure of

RE-1-silencing transcription factor (REST), a repressor of neuronal genes. J. Biol.

Chem 273, 26891-26899 (1998).

21. Tapia-Ramírez, J., Eggen, B., Peral-Rubio, M., Toledo-Aral, J. & Mandel, G. A

single zinc finger motif in the silencing factor REST represses the neural-specific

type II sodium channel  promoter. Proc Natl Acad Sci U S A. 94, 1177–1182

(1997).

22. Naruse, Y., Aoki, T., Kojima, T. & Mori, N. Neural restrictive silencer factor

recruits mSin3 and histone deacetylase complex to repress neuron-specific target

genes. Proceedings of the National Academy of Sciences of the United States of

America 96, 13691-13696 (1999).

23. Grimes, J.A. et al. The co-repressor mSin3A is a functional component of the

REST-CoREST repressor complex. J. Biol. Chem 275, 9461-9467 (2000).

24. Andrés, M.E. et al. CoREST: A functional corepressor required for regulation of

neural-specific gene expression. Proceedings of the National Academy of Sciences

of the United States of America 96, 9873-9878 (1999).

25. Roopra, A. et al. Transcriptional Repression by Neuron-Restrictive Silencer Factor

Is Mediated via the Sin3-Histone Deacetylase Complex. Mol. Cell. Biol. 20, 2147-

2157 (2000).

26. Huang, Y., Myers, S.J. & Dingledine, R. Transcriptional repression by REST:

recruitment of Sin3A and histone deacetylase to neuronal genes. Nat Neurosci 2,

867-872 (1999).

27. Jepsen, K. et al. Combinatorial Roles of the Nuclear Receptor Corepressor in

Transcription and Development. Cell 102, 753-763 (2000).

28. Lunyak, V.V. et al. Corepressor-Dependent Silencing of Chromosomal Regions

Encoding Neuronal Genes. Science 298, 1747-1752 (2002).

29. Ballas, N. & Mandel, G. The many faces of REST oversee epigenetic

programming of neuronal genes. Current Opinion in Neurobiology 15, 500-506

(2005).

30. Chen, Z., Paquette, A.J. & Anderson, D.J. NRSF/REST is required in vivo for

repression of multiple neuronal target genes during embryogenesis. Nat Genet 20,

136-142 (1998).

31. Jones, F.S. & Meech, R. Knockout of REST/NRSF shows that the protein is a

potent repressor of neuronally expressed genes in non-neural tissues. BioEssays

21, 372-376 (1999).

32. Jorgensen, H.F. et al. REST selectively represses a subset of RE1-containing

neuronal genes in mouse embryonic stem cells. Development 136, 715-721 (2009).

33. Singh, S.K., Kagalwala, M.N., Parker-Thornburg, J., Adams, H. & Majumder, S.

REST maintains self-renewal and pluripotency of embryonic stem cells. Nature

453, 223-227 (2008).

34. Yang, Y. et al. NRSF silencing induces neuronal differentiation of human

mesenchymal stem cells. Experimental Cell Research 314, 2257-2265

84

35. Immaneni, A. et al. REST-VP16 activates multiple neuronal differentiation genes

in human NT2 cells. Nucl. Acids Res. 28, 3403-3410 (2000).

36. Su, X., Kameoka, S., Lentz, S. & Majumder, S. Activation of REST/NRSF Target

Genes in Neural Stem Cells Is Sufficient To Cause Neuronal Differentiation. Mol

Cell Biol. 24, 8018–8025 (2004).

37. Schoenherr, C.J., Paquette, A.J. & Anderson, D.J. Identification of potential target

genes for the neuron-restrictive silencer factor. Proc Natl Acad Sci U S A. 93,

9881–9886 (1996).

38. Bruce, A.W. et al. Genome-wide analysis of repressor element 1 silencing

transcription factor/neuron-restrictive silencing factor (REST/NRSF) target genes.

Proc Natl Acad Sci U S A. 101, 10458–10463 (2004).

39. Palm, K., Belluardo, N., Metsis, M. & Timmusk, T.O. Neuronal Expression of

Zinc Finger Transcription Factor REST/NRSF/XBR Gene. J. Neurosci. 18, 1280-

1296 (1998).

40. Wood, I.C. et al. Interaction of the Repressor Element 1-silencing Transcription

Factor (REST) with Target Genes. Journal of Molecular Biology 334, 863-874

(2003).

41. Nishimura, E., Sasaki, K., Maruyama, K., Tsukada, T. & Yamaguchi, K. Decrease

in neuron-restrictive silencer factor (NRSF) mRNA levels during differentiation of

cultured neuroblastoma cells. Neuroscience Letters 211, 101-104 (1996).

42. Ballas, N., Grunseich, C., Lu, D.D., Speh, J.C. & Mandel, G. REST and its

corepressors mediate plasticity of neuronal gene chromatin throughout

neurogenesis. Cell 121, 645-657 (2005).

43. Kuwabara, T., Hsieh, J., Nakashima, K., Taira, K. & Gage, F.H. A Small

Modulatory dsRNA Specifies the Fate of Adult Neural Stem Cells. Cell 116, 779-

793 (2004).

44. Sun, Y. et al. Distinct Profiles of REST Interactions with Its Target Genes at

Different Stages of Neuronal Development. Mol. Biol. Cell 16, 5630-5638 (2005).

45. Paquette, A.J., Perez, S.E. & Anderson, D.J. Constitutive expression of the

neuron-restrictive silencer factor (NRSF)/REST in differentiating neurons disrupts

neuronal gene expression and causes axon pathfinding errors in vivo. Proceedings

of the National Academy of Sciences of the United States of America 97, 12318-

12323 (2000).

46. Ballas, N. et al. Regulation of neuronal traits by a novel transcriptional complex.

Neuron 31, 353-365 (2001).

47. Griffith, E.C., Cowan, C.W. & Greenberg, M.E. REST Acts through Multiple

Deacetylase Complexes. Neuron 31, 339-340 (2001).

48. Lepagnol-Bestel, A. et al. Nrsf silencing induces molecular and subcellular

changes linked to neuronal plasticity. Neuroreport 18, 441-446 (2007).

49. Bessis, A., Champtiaux, N., Chatelin, L. & Changeux, J. The neuron-restrictive

silencer element: A dual enhancer/silencer crucial for patterned expression of a

nicotinic receptor gene in the  brain. Proc Natl Acad Sci U S A. 94, 5906–5911

(1997).

50. Kallunki, P., Edelman, G.M. & Jones, F.S. The neural restrictive silencer element

can act as both a repressor and enhancer of L1 cell adhesion molecule gene

85

expression during postnatal  development. Proc Natl Acad Sci U S A. 95, 3233–

3238 (1998).

51. Seth, K.A. & Majzoub, J.A. Repressor Element Silencing Transcription

Factor/Neuron-restrictive Silencing Factor (REST/NRSF) Can Act as an Enhancer

as Well as a Repressor of Corticotropin-releasing Hormone Gene Transcription. J.

Biol. Chem. 276, 13917-13923 (2001).

52. Yoo, J. et al. The Neuron Restrictive Silencer Factor Can Act as an Activator for

Dynamin I Gene Promoter Activity in Neuronal Cells. Biochemical and

Biophysical Research Communications 283, 928-932 (2001).

53. Kim, C.S. et al. Novel function of neuron-restrictive silencer factor (NRSF) for

posttranscriptional regulation. Biochimica et Biophysica Acta (BBA) - Molecular

Cell Research 1783, 1835-1846 (2008).

54. Shimojo, M., Paquette, A.J., Anderson, D.J. & Hersh, L.B. Protein Kinase A

Regulates Cholinergic Gene Expression in PC12 Cells: REST4 Silences the

Silencing Activity of Neuron-Restrictive Silencer Factor/REST. Mol. Cell. Biol.

19, 6788-6795 (1999).

55. Abramovitz, L. et al. Dual role of NRSF/REST in activation and repression of the

glucocorticoid response. J. Biol. Chem 283, 110-119 (2008).

56. Majumder, S. REST in good times and bad: roles in tumor suppressor and

oncogenic activities. Cell Cycle 5, 1929-1935 (2006).

57. Westbrook, T.F. et al. A Genetic Screen for Candidate Tumor Suppressors

Identifies REST. Cell doi:10.1016/j.cell.2005.03.033

58. Lawinger, P. et al. The neuronal repressor REST/NRSF is an essential regulator in

medulloblastoma cells. Nat Med 6, 826-831 (2000).

59. Coulson, J.M. Transcriptional Regulation: Cancer, Neurons and the REST.

Current Biology 15, R665-R668 (2005).

60. Zuccato, C. et al. Huntingtin interacts with REST/NRSF to modulate the

transcription of NRSE-controlled neuronal genes. Nat Genet 35, 76-83 (2003).

61. Zuccato, C. et al. Widespread Disruption of Repressor Element-1 Silencing

Transcription Factor/Neuron-Restrictive Silencer Factor Occupancy at Its Target

Genes in Huntington's Disease. J. Neurosci. 27, 6972-6983 (2007).

62. Sempere, L.F. et al. Expression profiling of mammalian microRNAs uncovers a

subset of brain-expressed microRNAs with possible roles in murine and human

neuronal differentiation. Genome Biol. 5, R13 (2004).

63. Lewis, B.P., Burge, C.B. & Bartel, D.P. Conserved Seed Pairing, Often Flanked

by Adenosines, Indicates that Thousands of Human Genes are MicroRNA Targets.

Cell 120, 15-20 (2005).

64. Mortazavi, A., Leeper Thompson, E.C., Garcia, S.T., Myers, R.M. & Wold, B.

Comparative genomics modeling of the NRSF/REST repressor network: From

single conserved sites to genome-wide repertoire. Genome Res. 16, 1208–1221

(2006).

65. Johnson, D.S., Mortazavi, A., Myers, R.M. & Wold, B. Genome-Wide Mapping

of in Vivo Protein-DNA Interactions. Science 316, 1497-1502 (2007).

66. Liang, Y., Li, Q., Zhang, X., Shi, S. & Jing, G. Differential expression of nuclear

matrix proteins during the differentiation of human neuroblastoma SK-N-SH cells

86

induced by retinoic acid. J. Cell. Biochem. 106, 849-857 (2009).

67. Biedler, J.L., Helson, L. & Spengler, B.A. Morphology and growth,

tumorigenicity, and cytogenetics of human neuroblastoma cells in continuous

culture. Cancer Res 33, 2643-2652 (1973).

68. Rostomily, R.C. et al. Expression of neurogenic basic helix-loop-helix genes in

primitive neuroectodermal tumors. Cancer Res 57, 3526-3531 (1997).

69. Seeger, R.C. et al. Morphology, growth, chromosomal pattern and fibrinolytic

activity of two new human neuroblastoma cell lines. Cancer Res 37, 1364-1371

(1977).

70. Biedler, J.L., Roffler-Tarlov, S., Schachner, M. & Freedman, L.S. Multiple

neurotransmitter synthesis by human neuroblastoma cell lines and clones. Cancer

Res 38, 3751-3757 (1978).

71. Fults, D., Pedone, C.A., Morse, H.G., Rose, J.W. & McKay, R.D. Establishment

and characterization of a human primitive neuroectodermal tumor cell line from

the cerebral hemisphere. J. Neuropathol. Exp. Neurol 51, 272-280 (1992).

72. Myers, R.M. Tasks. HTSWorkflow HudsonAlpha at

<http://htsw.stanford.edu/admin/>

73. Zhang, Y. et al. Model-based Analysis of ChIP-Seq (MACS). Genome Biol. 9,

R137 (2008).

74. Galaxy. at <http://galaxy.psu.edu/>

75. Berriz, G. Funcassociate: The Gene Set Functionator. Roth Laboratory at

<http://llama.med.harvard.edu/cgi/func1/funcassociate_advanced>

76. Mortazavi, A., Williams, B.A., McCue, K., Schaeffer, L. & Wold, B. Mapping

and quantifying mammalian transcriptomes by RNA-Seq. Nat Meth 5, 621-628

(2008).

77. Bibel, M., Richter, J., Lacroix, E. & Barde, Y. Generation of a defined and

uniform population of CNS progenitors and neurons from mouse embryonic stem

cells. Nat. Protocols 2, 1034-1043 (2007).