Towards elucidating the stability, dynamics andarchitecture of the nucleosome remodeling anddeacetylase complex by using quantitative interactionproteomicsSusan L. Kloet1*, H. Irem Baymaz1*, Matthew Makowski1*, Vincent Groenewold2, Pascal W. T. C.Jansen1, Madeleine Berendsen1, Hassin Niazi1, Geert J. Kops2,3 and Michiel Vermeulen1,2,3

1 Department of Molecular Biology, Faculty of Science, Radboud Institute for Molecular Life Sciences, Radboud University Nijmegen,

the Netherlands

2 Department of Molecular Cancer Research, UMC Utrecht, the Netherlands

3 Cancer GenomiCs Netherlands, the Netherlands

Keywords

chromatin; mass spectrometry; NuRD;

protein-protein interactions; quantitative

proteomics

Correspondence

M. Vermeulen, Department of Molecular

Biology, M850/3.79 259 RIMLS, P.O. Box

9101 6500 HB Nijmegen, The Netherlands

Tel: +31 024 3616848

E-mail: m.vermeulen@ncmls.ru.nl

*These authors contributed equally to this

work.

(Received 15 July 2014, revised 7 August

2014, accepted 12 August 2014)

doi:10.1111/febs.12972

The nucleosome remodeling and deacetylase (NuRD) complex is an evolu-

tionarily conserved chromatin-associated protein complex. Although the sub-

unit composition of the mammalian complex is fairly well characterized, less

is known about the stability and dynamics of these interactions. Furthermore,

detailed information regarding protein–protein interaction surfaces within the

complex is still largely lacking. Here, we show that the NuRD complex inter-

acts with a number of substoichiometric zinc finger-containing proteins. Some

of these interactions are salt-sensitive (ZNF512B and SALL4), whereas others

(ZMYND8) are not. The stoichiometry of the core subunits is not affected

by high salt concentrations, indicating that the core complex is stabilized by

hydrophobic interactions. Interestingly, the RBBP4 and RBBP7 proteins are

sensitive to high nonionic detergent concentrations during affinity purifica-

tion. In a subunit exchange assay with stable isotope labeling by amino acids

in cell culture (SILAC)-treated nuclear extracts, RBBP4 and RBBP7 were

identified as dynamic core subunits of the NuRD complex, consistent with

their proposed role as histone chaperones. Finally, using cross-linking MS,

we have uncovered novel features of NuRD molecular architecture that com-

plement our affinity purification-MS/MS data. Altogether, these findings

extend our understanding of MBD3–NuRD structure and stability.

Structured digital abstract

� MBD3 physically interacts with ZNF512B, HDAC1, ZMYND8, GATAD2B, SALL4,

GATAD2A, ZNF592, MTA3, ZNF687, CDK2AP1, CHD3, ZNF532, HDAC2, MTA2,

CHD4, MTA1, KPNA2, CHD5, RBBP4 and RBBP7 by pull down (View interaction)

� CDK2AP1 physically interacts with MBD3, MTA3, HDAC2, GATAD2A, CHD4,

CDK2AP1, MTA2, HDAC1, MTA1, CHD3, GATAD2B, MBD2,

RBBP4 and RBBP7 by pull down (View interaction)

� MBD3 physically interacts with MTA2, MTA3, RBBP4, RBBP7, HDAC2, HDAC1,

CHD4, CHD3 and MTA1 by cross-linking study (View interaction)

Abbreviations

AP, affinity purification; FDR, false discovery rate; GFP, green fluorescent protein; HDAC, histone deacetylase; iBAQ, intensity-based

absolute quantification; NuRD, nucleosome remodeling and deacetylase; PDB, Protein Data Bank; SAS, solvent-accessible surface; SILAC,

stable isotope labeling by amino acids in cell culture; TAP, tandem affinity purification; TEV, tobacco etch virus.

1FEBS Journal (2014) ª 2014 FEBS

Introduction

The nucleosome remodeling and deacetylase (NuRD)

complex is a conserved chromatin-associated protein

complex that was first purified and characterized from

Xenopus laevis and human cells in the late 1990s [1–3].As the name implies, the NuRD complex has both

histone deacetylase (HDAC) and ATP-dependent

chromatin remodeling activity. In the human complex,

HDAC activity is catalyzed by HDAC1 and HDAC2.

The large CHD3 and CHD4 subunits have ATPase

activity, which is used to move the position of nucleo-

somes on DNA. In addition to these enzymatic activi-

ties, the NuRD complex contains a number of other

core subunits: MTA1–MTA3, GATAD2A, GA-

TAD2B, MBD2, MBD3, RBBP4, RBBP7, and

CDK2AP1 [4,5]. Some of these proteins, such as

MBD2 and MBD3, are mutually exclusive within the

NuRD complex. For GATAD2A, GATAD2B,

RBBP4, RBBP7, and MTA1–MTA3, it is currently

not known whether these proteins exclusively form

heterodimers/trimers or form mutually exclusive

homodimers/trimers. In addition to the well-described

core subunits, a large number of proteins have been

reported to interact with the NuRD complex. Exam-

ples include SALL4, FOG1, and ZMYND8 [6–8].These proteins may recruit the NuRD complex to its

target sites in the genome.

Owing to the presence of HDACs, which are gener-

ally associated with gene silencing, the NuRD complex

has long been thought of as a transcriptional corepres-

sor complex. Indeed, in luciferase assays, NuRD

subunits repress the transcription of reporter constructs

[4]. Furthermore, the methyl-CpG-binding protein

MBD2 links MBD2-containing NuRD to methylated

CpG islands, and induces gene silencing [9]. However,

recent genome-wide profiling studies have revealed that

NuRD complexes, such as MBD3–NuRD, are also

found at active enhancers and promoters, indicating

that NuRD-mediated regulation of transcription is

more diverse than previously thought [10,11]. Also,

functional experiments have revealed an important role

for the NuRD complex in regulating cell fate decisions

[12,13]. These different biological functions of the

NuRD complex may be driven by subtle changes in

subunit composition.

Recently, a number of MS-based methods were

developed that can be used not only to confidently

identify protein–protein interactions, but also to obtain

information about the stoichiometry, dynamics and

the interaction surfaces between subunits [14–16].Here, we applied these different methods to gain fur-

ther insights into the structure and dynamics of the

MBD3–NuRD complex in HeLa cells. We applied

label-free quantitative MS to investigate the stoichiom-

etry of the MBD3–NuRD complex, using different

buffer stringencies during affinity purification (AP). A

stable isotope labeling by amino acids in cell culture

(SILAC)-based subunit exchange assay was used to

show that RBBP4 and RBBP7 are dynamic core su-

bunits within the NuRD complex. Finally, cross-link-

ing MS was used to identify interprotein contacts

within the MBD3–NuRD complex. Altogether, these

experiments increase our understanding of the stabil-

ity, dynamics and architecture of the NuRD complex.

Results

RBBP4 and RBBP7 interactions with the NuRD

complex are salt-stable but detergent-sensitive

To facilitate purification of the MBD3–NuRD com-

plex, we created a stable cell line with doxycycline-

inducible green fluorescent protein (GFP)-tagged

MBD3. Cells were incubated for 16 h in the presence

or absence of doxycycline. Expression of GFP–MBD3

occurred only after addition of doxycycline to the

medium (Fig. 1A). Nuclear extracts from this cell line

were used for GFP AP combined with label-free LC-

MS/MS analysis. Briefly, triplicate pulldowns were

performed with GFP-Trap_A beads and GFP–MBD3-

containing nuclear extracts (+Dox). Simultaneously,

triplicate pulldowns were performed with GFP beads

on control nuclear extracts (�Dox). Proteins that bind

specifically to GFP–MBD3 will be enriched in the

GFP pulldowns and appear on the right side of the

Fig. 1. Label-free proteomics reveal that RBBP4 and RBBP7 interactions with NuRD are NP-40-sensitive. (A) An anti-MBD3 western blot of

GFP–MBD3 HeLa cells after 16 h of incubation with doxycycline (+Dox) or without doxycycline (�Dox). Nonspecific bands are marked with

an asterisk. (B–D) Volcano plots from label-free GFP pulldowns of GFP–MBD3 HeLa cell nuclear extracts with varying salt and NP-40

concentrations. Statistically enriched proteins in the GFP–MBD3 pulldown are identified by a permutation-based FDR-corrected t-test. The

label-free quantification (LFQ) intensity of the GFP pulldown relative to the control [fold change (FC), x-axis] is plotted against the – log10-

transformed P-value of the t-test (y-axis). The proteins in the upper right corner represent the bait and its interactors. (E) Stoichiometry of

NuRD core subunits. The iBAQ value of each protein group is divided by the iBAQ value of the GFP–MBD3 bait, which was set to 2. (F)

Stoichiometry of NuRD complex interactors relative to the GFP–MBD3 bait determined with the same data as in (E), but with a differently

scaled axis.

2 FEBS Journal (2014) ª 2014 FEBS

Dynamics and architecture of the NuRD complex S. L. Kloet et al.

–15 –10 –5 0 5 10 15

01

23

45

67

Log2 (GFP/Control)

– Lo

g [F

DR

(t-te

st)]

GATAD2ACHD3

CDK2AP1

MTA2

MBD3

KPNA2

RBBP4

MTA1

HDAC1

CHD4

ZNF687

CHD5

GATAD2B

ZNF592

HDAC2

ZNF512BMTA3SALL4

FC > 3.96 FDR > 1.301

300 mM NaCl 0.1% NP-40

RBBP7

ZMYND8

–15 –10 –5 0 5 10 15

02

46

Log2 (GFP/Control)

– Lo

g [F

DR

(t-te

st)]

GATAD2A

CHD3

CDK2AP1 MTA2MBD3

KPNA2

RBBP4

MTA1

HDAC1

CHD4

ZNF687

CHD5

GATAD2B

ZNF592 HDAC2

ZNF512B

MTA3

SALL4

FC > 4.65 FDR > 1.301

300 mM NaCl0.5% NP-40

RBBP7

ZMYND8

–15 –10 –5 0 5 10 15

01

23

45

67

Log2 (GFP/Control)

– Lo

g [F

DR (t

-test

)]

GATAD2A

CHD3

CDK2AP1

MTA2

MBD3KPNA2

RBBP4

MTA1

HDAC1

CHD4

ZNF687

CHD5 GATAD2B

ZNF592HDAC2

ZNF512B

MTA3

SALL4

FC > 4.04 FDR > 1.301

1 M NaCl0.5% NP-40

ZMYND8

RBBP7

13

57

– Dox + DoxGFP-MBD3*

*

*MBD3

A

B

C

D

E

α-MBD3

F

Stoichiometry

Stoichiometry

3FEBS Journal (2014) ª 2014 FEBS

S. L. Kloet et al. Dynamics and architecture of the NuRD complex

plot. Proteins that bind nonspecifically to the GFP

beads will appear in the background cloud.

To determine the stability of the NuRD complex,

several NP-40 and NaCl concentrations were used dur-

ing the APs (Fig. 1B–D). In each of these pulldowns,

known NuRD core subunits were identified. In addi-

tion, several previously described NuRD interactors

were detected, including ZMYND8, ZNF592, and

SALL4 [8,14,17]. Interestingly, we did not pull down

LSD1 in any of our GFP–MBD3 purifications. LSD1

has previously been reported as a putative subunit of

the NuRD complex [18], but our data on the MBD3–NuRD complex do not agree with this observation.

Next, the intensity-based absolute quantification (iBAQ)

values were compared between all NuRD subunits and

interactors. The iBAQ algorithm normalizes the total

MS intensity for each protein according to the number

of theoretically observable peptides [19]. This allows

estimation of the relative abundance of large and small

proteins detected in APs. As the NuRD complex con-

tains many paralogs that overlap at the peptide level,

iBAQ values were summed for all paralogs (Table S1).

Comparison of the iBAQ values relative to the GFP–MBD3 bait revealed that most core subunits of

NuRD remained tightly bound to each other and to

MBD3 despite the presence of high salt (1 M NaCl)

and NP-40 (0.5%) concentrations in the wash steps

(Fig. 1E). Surprisingly, the core RBBP4 and RBBP7

subunits were very sensitive to NP-40. Increasing the

NP-40 concentration from 0.15% to 0.5% led to

decreases in the amounts of RBBP4 and RBBP7 asso-

ciated with the complex (stoichiometry of 5.5 reduced

to 4).

In addition, the association of several zinc finger

domain-containing proteins with NuRD was very sen-

sitive to high salt concentrations. The association of

ZNF512B decreased nearly eight-fold when the salt

concentration in the wash was increased from 300 mM

to 1 M NaCl (Fig. 1F). A reduction in SALL4 binding

was also observed with increased salt concentration. In

contrast, other substoichiometric interactors, such as

ZMYND8 and KPNA2, remained tightly associated

with the core complex under increased salt and NP-40

concentrations. Together, these results suggest that the

core NuRD complex is very stable when challenged

with high salt concentrations, indicating that interac-

tions within the NuRD complex are at least partially

driven by hydrophobic interactions. The partial disso-

ciation of RBBP4 and RBBP7 observed at high NP-40

concentrations may suggest that these proteins are less

tightly associated with the rest of the core complex.

Finally, although other detected interactors had a low

iBAQ value relative to the core complex (1–2% rela-

tive to the core subunits), some of these interactions

were very stable at high salt and NP-40 concentra-

tions, suggesting a functional role for these proteins in

relation to the NuRD complex.

RBBP4 and RBBP7 partially dissociate from the

NuRD complex in a subunit exchange assay

To further study the potential dynamics of NuRD sub-

unit interactions, a SILAC-based subunit exchange

assay was used (Fig. 2A) [15]. HeLa cells expressing

GFP–CDK2AP1 were labeled in culture with heavy

amino acids (forward) or light amino acids (reverse).

Similarly, wild-type HeLa cells were labeled with light

or heavy amino acids. GFP pulldowns were performed

immediately after mixing of the nuclear extracts (T0)

or after overnight incubation (TON). Proteins that are

more dynamically associated with NuRD will dissoci-

ate from the complex during the overnight incubation

step and may be replaced by proteins from the other,

differentially labeled extract. This eventually results in

a decrease in detected SILAC ratios, whereby dynamic

core subunits and/or interactors will move towards the

background cloud in the scatter plot.

At T0, all NuRD core subunits were significantly

enriched according to boxplot statistics (Fig. 2B).

However, after overnight incubation, RBBP4 and

RBBP7 clearly separated from the other NuRD subun-

its, and migrated towards the background cloud

(Fig. 2C). To more directly compare the two scatter

plots, the difference in forward and reverse ratios

between the plots was visualized in a graph (Fig. 2D).

A protein with no change in ratios between experi-

ments would have a value of 0. This graph clearly

shows that RBBP4 and RBBP7 are the most dynamic

NuRD core subunits. These observations are in agree-

ment with recent structural studies suggesting that

MTA and histone H4 compete for RBBP binding [20].

Furthermore, RBBP4 and RBBP7 are present in many

protein complexes other than the NuRD complex

(Sin3 complex, PRC2), which also could explain their

observed dynamic behavior [21,22].

Cross-linking MS reveals novel architectural

features of the human NuRD complex

To further characterize the structural interactions

between NuRD complex core subunits, we conducted

cross-linking MS to identify interprotein interaction

surfaces (Fig. 3A). The NuRD complex was isolated

from HeLa nuclear extracts by the use of tandem AP

with a GFP–His-MBD3 bait, and subsequently cross-

linked with BS3 for LC-MS/MS analysis. Cross-linking

4 FEBS Journal (2014) ª 2014 FEBS

Dynamics and architecture of the NuRD complex S. L. Kloet et al.

was monitored by silver staining to ensure complete

cross-linking of all NuRD subunits (Fig. 3B). High-

confidence cross-linked peptides [false discovery rate

(FDR) of < 0.05] were then identified with PLINK by

using default settings and searching against a database

of NuRD core components (Fig. 3C,D) [23]. In total,

336 spectra matching with high confidence to 16 non-

ambiguous interprotein cross-links, 87 nonambiguous

0 2 4 6

0–2

–4–6

–8–1

0

Forward H/L (log2)

Rev

erse

H/L

(log

2)

HDAC1

MBD2

HDAC2

MTA1

CDK2AP1

MTA3

MTA2

CHD3

CHD4

GATAD2A

MBD3

GATAD2B

RBBP4

RBBP7

T0

core

transient

core

transient

Forward

Reverse

Light WT

Heavy GFPMix

Incubation for 0 h (T0) or

Overnight (TON)

10-min GFP pulldown

Tryptic digest,LC-MS/MS

Heavy WT

Light GFPMix

T0

TON

ReverseForwardA

B

D

1 3 5 0 1 2 3 4 5 6

0–2

–4–6

–8

Forward H/L (log2)

Rev

erse

H/L

(log

2)

HDAC2HDAC1

MTA1

MTA3

MTA2

MBD3

CHD3

CHD4

MBD2GATAD2A

GATAD2B

RBBP4RBBP7

TON

CDK2AP1

C

Fig. 2. RBBP4 and RBBP7 are the most dynamic NuRD core subunits. (A) Schematic representation of the SILAC subunit exchange assay.

(B, C) Scatter plots of GFP–CDK2AP1 forward and reverse pulldowns immediately after extract mixing [(B), T0] and after overnight

incubation [(C), TON]. The SILAC ratio for each protein in the forward pulldown is plotted against the ratio in the reverse pulldown. Boxplot

statistics were used to determine statistically significant GFP–CDK2AP1 interactors. The arrow and circle in (C) illustrate the movement of

RBBP4 and RBBP7 towards the background cloud, indicating that these proteins exchange between the heavy (H) and light (L) extracts. (D)

Bar graph showing the difference in forward and reverse SILAC ratios between T0 and TON. WT, wild type.

5FEBS Journal (2014) ª 2014 FEBS

S. L. Kloet et al. Dynamics and architecture of the NuRD complex

MTA2

MBD3

RBBP4

CHD3 HDAC1

CHD4

RBBP7GATAD2A

GATAD2B

HDAC2

MTA1 MTA3

F

A

Hela cells

GFP-His MBD3

Collect cells

Extract nuclear proteins

GFPGFP

beadsMBD3

On bead cross-linking

MBD3

Enzymatic digestion

Cross-linksLoop-links

Mono-links

LC-MS/MS analysis

b3

b4

b5

b6

y2y3

y4

y5

validation of cross-linked peptides

P HVK WT R

Q VFKL Kb3 b5b4

y2

y3 y4 y5

b6

b7

b7y1

y6

y6

y1

B

TEV protease cleavage

MBD3

His

HisNi2+–NTA

beads

Ni2+–NTAbeads

C

200

150

100

50

0151050 2520 30

FDR (%)

Num

ber o

f spe

ctra

Estimated FDR curve

Charge 3+Charge 4+

His

58

46

80

175

RBBP4/7

CHD3/4

D

E

180°

y1+

y2+

y3+

y4+

y2+

y3+

y4+

y5+

y5+

y6+

y8+

y7+

+2b

+4b

+3b

00

60

50

40

30

20

10

100

90

80

70Re

lativ

e in

tens

ity (%

)

200 300 400 500 600 700 800 900

m/z1000

N SPM TK E

L VDAK Fy5

y8 y7 y6

b2 b3

y4

y5

L VI Ry4 y3 y2 y1

y3 y2

b4

K

HDAC1/2

MBD3

MTA1/2/3

TEVelute

Ni2+ - NTA

purificationsupernatant

Cross-linkingFraction

GFP bead purification

and Ni2+–nitrilotriacetic acidpurification

6 FEBS Journal (2014) ª 2014 FEBS

Dynamics and architecture of the NuRD complex S. L. Kloet et al.

intraprotein cross-links and 46 ambiguous cross-links

encompassing a variety of homomers or heteromers or

intraprotein cross-links were identified (Table S2).

Structural validation was confounded by a lack of

overlap between the few existing NuRD crystal struc-

tures and our identified cross-links. For example,

although five HDAC1 and 11 HDAC2 intraprotein

cross-links and four ambiguous HDAC1/HDAC2

cross-links were identified, all of them were located in

the undetermined region of the known HDAC1 crystal

structure [Protein Data Bank (PDB) 4BKX]. However,

seven cross-links mapping to the CHD4 chromodo-

main-ATPase region (residues 500–1298), which has a

yeast CHD1 ortholog with known structure (PDB

3MWY), were identified. To our knowledge, this

represents the highest number of mappable cross-links

from our dataset to any single structure. Structural

validation of our CHD4 chromodomain-ATPase cross-

links was performed with a homology modeling

approach. The human CHD4 sequence was aligned to

the yeast CHD1 template with HHPRED, and homology

models were constructed with MODELLER (Human

CHD4 Uniprot: Q14839) [24–27]. In 10 overlapping

homology models, well-ordered regions from the crys-

tal structure yielded consistently low divergence mod-

els, whereas short alignment gaps or structurally

undetermined loops were clearly disordered (Fig. 3E).

All homology models were highly similar, except for

very small, flexible regions, so one model was chosen

as representative, and cross-links were validated by

calculating the solvent-accessible surface (SAS) dis-

tance with XWALK [28]. All seven cross-links showed an

SAS distance of < 30 �A, as expected from the maxi-

mum length of the BS3 cross-linker. Three cross-links

mapped to a well-ordered region, and four mapped to

a disordered region that probably possesses a highly

flexible structural conformation. Therefore, all seven

CHD4 chromodomain-ATPase cross-links seem to be

structurally plausible, and we conclude that our cross-

linking data are of good quality.

Few NuRD subcomplexes have resolved structures,

most notably the MBD2–GATAD2A coiled-coil sub-

complex, the RBBP4–MTA1 subcomplex, and the

HDAC1–MTA1 subcomplex [20,29,30]. None of these

studies described comprehensive protein structures for

both binding partners, so the complete interaction

interfaces remain unknown. Therefore, our interpro-

tein cross-linking data reveal novel features of the

NuRD complex architecture and extend our under-

standing of NuRD structure–function relationships

(Fig. 3F). For example, robust contacts were observed

between MTA1–MTA3 and RBBP4 and RBBP7

subunits. Our data do not show whether these RBBP4

and RBBP7 cross-link sites indicate binding of a single

MTA paralog or a multimer, nor do they enable us to

distinguish between potentially dynamic or paralog-

specific interactions at each site. However, mapping

cross-linked residues onto the known RBBP4 structure

shows that Lys25 and Lys307 fall on either side of the

known MTA1 interaction interface (PDB 4PC0 [20])

(Fig. 4A). As RBBP4 Lys25 maps relatively close to

the known MTA1-binding pocket, whereas Lys307 sits

across a large, adjacent hydrophobic region, these

cross-links could outline the opposite ends of the

MTA1–MTA3 binding platform (Fig. 4B). Intrigu-

ingly, this would explain our observation that MTA1–MTA3 binding to RBBP4 and RBBP7 is stabilized by

hydrophobic interactions under high-salt conditions.

Similarly, although the MBD3 structure is not known,

we identified a short MBD3 region (residues 92–124)that contained five cross-links with GATAD2A/GA-

TAD2B, two of which represented unique cross-links

for GATAD2A, one that was unique for GATAD2B,

and two that were ambiguous for the GATAD2A/GA-

Fig. 3. Cross-linking MS offers insight into the NuRD molecular architecture. (A) MBD3–NuRD was obtained by tandem AP (GFP affinity

enrichment, TEV cleavage, and Ni2+–nitrilotriacetic acid (NTA) enrichment) and cross-linked on-bead with BS3. Proteolytic digestion and

analysis by AP-MS/MS was followed by identification and validation of high-confidence cross-linked peptides (FDR < 0.05) with PLINK. (B)

NuRD complex purification and cross-linking was monitored by silver staining to ensure complete cross-linking of all complex components.

Note that, after the cross-linking reaction, NuRD subunits migrate at an apparent higher molecular mass in the SDS/PAGE gel. (C) High-

confidence spectra were matched to putative cross-links by use of a target-decoy approach in PLINK. Representative FDR curves are shown

for cross-links of charge states +3 and +4. (D) An annotated high-scoring spectrum from a CHD4 intraprotein cross-link, Lys959–Lys969. (E)

Structural validation was performed with a human CHD4 chromodomain-ATPase homology model aligned to a yeast CHD1 ortholog

template structure. Yeast CHD1 is colored red, with 10 overlapping CHD4 homology models in blue. Validated cross-links are colored

purple. Below the homology models are close-up views of seven CHD4 cross-links in the chromodomain-ATPase region obtained with a

representative CHD4 homology model; the lower left image shows Lys959–Lys969, Lys959–Lys1189 and Lys969–Lys1189 cross-links in a

well-ordered region, and the lower right image shows Lys690–Lys693 and Lys693–Lys696 cross-links in a gapped region, and Lys833–

Lys838 and Lys833–Lys839 cross-links in a structurally undetermined region. (F) Cross-link map for human NuRD subunits. Subunits

connected by at least one unambiguous cross-link are indicated by solid lines, and subunits connected only by ambiguous cross-links are

indicated by dashed lines. In our data, all ambiguous cross-links are ambiguous for peptides from a paralogous subunit. The relative

numbers of interprotein and intraprotein cross-links are designated by the area of gray shading surrounding that link.

7FEBS Journal (2014) ª 2014 FEBS

S. L. Kloet et al. Dynamics and architecture of the NuRD complex

TAD2B peptide (Fig. 4C). The MBD2 and MBD3

sequences were aligned in EBI’s CLUSTAL OMEGA in

order to map any MBD3–GATAD2A cross-links with

the solved MBD2–GATAD2A coiled-coil structure

(PDB 2L2L [30,31]). Only two cross-linked lysines

could be mapped to a representative conformer GA-

TAD2A structure (Lys163 and Lys178), and no cross-

links could be mapped to the MBD2 structure

(Fig. 4D). Furthermore, one GATAD2A cross-link

site, Lys178, was located at the tail end of the con-

former, which showed a highly flexible conformation

and thus offers little structural information. It is possi-

ble that we observed low cross-link coverage within

the MBD2–GATAD2A coiled-coil domain because, as

the authors of that study noted, this interaction hides

at least 1337 �A2 of the solvent-accessible area, poten-

tially hindering BS3 accessibility. Thus, our cross-links

may outline an interaction interface rather than reveal

a binding footprint. However, in agreement with previ-

ous work, our data show a single binding interface

between MBD3 and GATAD2A/GATAD2B paralogs.

Overall, these cross-linking MS data greatly improve

MBD3BD331 200 291

GATAD2A

1 200 400 633

GATAD2B

1 200 400 593

Coiled-coildomain

Coiled-coildomain

MTA2A21 200 400 600 668

MTA3A31 200 400 594

MTA1A11 200 400 600 715

RBBP4P41 200 425

RBBP7P71 200 425

RBBP4 interaction domain

RBBP4 crystalstructure

90°

GATAD2A-K163

RBBP4-K307

RBBP4-K25

GATAD2A-K178

MTA1 Fragment/H4 Binding pocket (MTA1 Fragment Removed)

Uncharacterized RBBP4 hydrophobic pocket

C D

A

B

Fig. 4. Structural analysis of cross-linked residues reveals novel features of NuRD subcomplex interactions. (A) Twelve identified cross-links

between RBBP4 and RBBP7 and MTA1–MTA3 map to two residues on RBBP4, i.e. K25 and K307 (K24 and K306 on RBBP7). (B) Residues

from (A) are mapped onto the known crystal structure for RBBP4. These residues surround the binding site of the MTA1 peptide. RBBP4-

K25 sits near the MTA1-binding site, whereas RBBP4-K307 lies across a large hydrophobic region representing a potential MTA1-binding

platform. Hydrophobic regions are colored red, and hydrophilic regions are colored blue. (C) Five cross-links identified between MBD3 and

GATAD2A/GATAD2B map to MBD3 residues K92, K109, and K124. (D) Coiled-coil conformer model of GATAD2A and MBD2. The

GATAD2A coil is colored light red, and the MBD2 coil is colored blue. All lysines are colored dark red, and the cross-linked residues are

colored yellow.

8 FEBS Journal (2014) ª 2014 FEBS

Dynamics and architecture of the NuRD complex S. L. Kloet et al.

our understanding of the NuRD complex subunit

architecture, and will help to direct future structural

studies.

Discussion

Here, we have used a variety of MS-based methods to

increase our understanding of the dynamics, stability

and architecture of the MBD3–NuRD complex. GFP-

based APs combined with a label-free method to esti-

mate protein complex stoichiometry revealed that the

RBBP4 and RBBP7 core subunits of the NuRD com-

plex are sensitive to high detergent concentrations.

Increasing the NP-40 concentration from 0.1% to 0.5%

in the wash buffer resulted in a reduced amount of

RBBP4 and RBBP7 copurified with MBD3. This find-

ing illustrates an important limitation of AP-MS experi-

ments, namely that some of the interactions that occur

in vivo may be lost during AP. Conversely, interactions

can be formed in the extract that may not occur in vivo.

To overcome these limitations, cross-linking prior to

cell lysis can be used as an alternative approach. How-

ever, this approach is technically very challenging,

owing to experimental and computational issues.

Except for MBD3 and HDAC1/HDAC2, the stoichi-

ometry values that we obtained in our GFP–MBD3

purifications are generally in good agreement with our

previously published NuRD stoichiometry results

[14,32]. Previously, we made use of a BAC GFP–MBD3 HeLa cell line that expresses MBD3 with a

GFP tag at near-endogenous levels. In the current

study, we used a doxycycline-inducible GFP–MBD3

cell line. This cell line may express MBD3 at slightly

higher levels, which may result in different MBD3 and

HDAC1/HDAC2 stoichiometry values. It is important

to note that the final stoichiometry values that we

obtained represent an average value based on a large

number of potentially heterogeneous complexes. On the

basis of our data, we cannot be sure whether paralo-

gous proteins such as MTA1–MTA3 or GATAD2A/

GATAD2B are present within the same complex, or

whether these proteins are, in fact, mutually exclusive,

as we have previously shown for MBD2 and MBD3

[33]. It is highly likely that a large number of distinct

NuRD complexes exist. However, our cross-linking

data cannot distinguish unambiguously between homo-

meric and heteromeric interactions for HDAC1/

HDAC2, MTA1–MTA3 and GATAD2A/GATAD2B.

Interestingly, we were able to detect unambiguous inter-

protein cross-links for CHD3 and CHD4, indicating

that these proteins may form a heterodimer. Neverthe-

less, NuRD complex heterogeneity most likely creates a

significant challenge for high-resolution determination

of the NuRD molecular architecture.

In addition to the core NuRD subunits, a number

of substoichiometric interactors were identified in the

GFP–MBD3 purifications. Interestingly, these interac-

tors showed differential sensitivity to salt and deter-

gent. ZMYND8 and ZNF532/592/687 copurified with

GFP–MBD3 even when challenged with 1 M NaCl

and 0.5% NP-40. The stability of these interactions

indicates that they may be functionally relevant. Inter-

estingly, these proteins have previously been reported

to interact with each other as a central hub in a large

transcriptional network [34]. Given the presence of a

large number of putative DNA-binding zinc fingers in

the ZMYND8/ZNF module, we hypothesize that this

module may recruit the NuRD complex to a subset of

its target genes. The same holds true for other substoi-

chiometric interactors. Further work is needed to

determine the functional significance of the ZMYND8/

ZNF–NuRD interaction.

Experimental procedures

Cloning

A tandem AP (TAP) tag was cloned into pcDNA5 FRT/

TO. This tag consists of enhanced GFP followed by two

tobacco etch virus (TEV) protease cleavage sites and a His6tag. The TEV sites and His tag were added to enhanced

GFP via PCR. MBD3 was amplified with BamHI and

XhoI, and cloned into the multiple cloning site of

pcDNA5 FRT/TO.

Cell culture and SILAC

HeLa FRT cells were transfected with pcDNA5 FRT/TO

TAP–MBD3 by the use of Lipofectamine LTX Plus (Invi-

trogen, Glasgow, Scotland). Cells underwent selection for

10 days with hygromycin, and single colonies were picked,

expanded, and screened for expression of TAP–MBD3.

TAP–MBD3 expression was induced by treatment with

1 lg�mL�1 doxycycline for 16 h. SILAC treatment of HeLa

cells was performed as described previously [35].

Nuclear extracts

Nuclear extracts were prepared essentially according to Dig-

nam et al. [36]. Cells were harvested with trypsin, washed

twice with PBS, and centrifuged at 400 g for 5 min at 4 °C.Cells were allowed to swell for 10 min at 4 °C in five volumes

of Buffer A (10 mM Hepes/KOH, pH 7.9, 1.5 mM MgCl2,

10 mM KCl), and then pelleted at 400 g for 5 min at 4 °C.Cells were resuspended in two volumes of Buffer A plus pro-

9FEBS Journal (2014) ª 2014 FEBS

S. L. Kloet et al. Dynamics and architecture of the NuRD complex

tease inhibitors and 0.15% NP-40, and transferred to a Do-

unce homogenizer. After 30–40 strokes with a Type B pestle,

the lysates were spun at 3200 g for 15 min at 4 °C. The

nuclear pellet was washed once with PBS, and spun at 3200 g

for 5 min at 4 °C. The nuclear pellet was resuspended in

Buffer C (420 mM NaCl, 20 mM Hepes/KOH, pH 7.9,

20% v/v glycerol, 2 mM MgCl2, 0.2 mM EDTA) with 0.1%

NP-40, protease inhibitors, and 0.5 mM dithiothreitol. The

suspension was incubated with rotation for 1 h at 4 °C, andthen spun at 18 000 g for 15 min at 4 °C. The supernatant

was saved, aliquoted, and stored at �80 °C until further use.

Label-free pulldowns and cross-linking MS

Label-free GFP pulldowns, LC-MS/MS and data analysis

were performed as described previously [14]. For cross-link-

ing MS, 8 mL of HeLa nuclear extract (5 mg�mL�1 plus

50 lm�mL�1 ethidium bromide) containing doxycycline-

induced TAP–MBD3 was incubated with 400 lL of GFP-

trap beads (Chromotek) for 2 h at 4 °C in a rotation

wheel. Beads were then washed three times with 10 mL of

wash buffer (300 mM NaCl, 0.5% NP-40, 20 mM Hepes/

KOH, pH 7.9, 0.5 mM dithiothreitol, 2 mM MgCl2, 0.2 mM

EDTA and complete protease inhibitors) and twice with

5 mL of TEV cleavage buffer (150 mM NaCl, 10 mM Tris/

HCl, pH 8.0, 0.1% NP-40, and 1 mM dithiothreitol). Beads

were then resuspended in 0.5 mL of TEV cleavage buffer

containing 100 units of GST–TEV protease (US Biological)

and incubated overnight at 4 °C in a rotation wheel. The

supernatant was collected, and GST–TEV protease was

removed from the eluate by adding 10 lL of GST beads

and incubating for 15 min at 4 °C in a rotation wheel. The

eluate was then incubated with 20 lL of Ni2+–nitrilotriace-

tic acid beads (Qiagen) for 45 min at 4 °C in a rotation

wheel. Beads were washed once with 1 mL of TEV cleav-

age buffer and three times with 1 mL of cross-linking buf-

fer (25 mM Hepes/KOH, pH 7.9, 300 mM KCl, 5%

glycerol, 0.05% Tween-20). Beads were then resuspended in

50 lL of cross-linking buffer containing 250 lM BS3 cross-

linker (Sigma), and incubated for 30 min at 37 °C in a

thermoshaker (1100 r.p.m.). The cross-linking reaction was

quenched by adding 7 lL of 0.5 M ammonium bicarbonate

and incubating for 5 min at room temperature. Finally,

beads were washed twice with 0.5 mL of PBS, after which

bound cross-linked proteins were digested overnight with

on-bead LysC/trypsin digestion. Tryptic peptides were de-

salted and purified with stagetips prior to LC-MS/MS

analysis.

Analysis of cross-linking MS data

Thermo RAW MS files were converted to mgf format with

the MSCONVERT tool from PROTEOWIZARD, and analyzed with

PLINK, with default settings for HCD with BS3 as a cross-

linking agent [23,37]. Cross-linked peptides matched to

spectra with an FDR of < 0.05 were considered to be high

confidence, and were used for further analysis. Spectra were

visually analyzed and annotated with PLABEL [38]. Template

structures for homology modeling were searched with

HHPRED, and homology models were built with the MODEL-

LER interface in UCSF CHIMERA [24,26,39]. The yeast CHD1

chromodomain and ATPase residues 38–800 (PDB 3MWY)

were used as a template for residues 500–1298 from human

CHD4 (Uniprot Q14839) for homology modeling [27].

Calculations of rmsd values were performed in UCSF CHI-

MERA. SAS distances were calculated for a representative

CHD4 model with XWALK. XINET was used to visualize

high-confidence intraprotein and interprotein cross-links in

network form (Rappsilber Laboratory, University of Edin-

burgh, Scotland; crosslinkviewer.org). All protein structure

visualizations were produced with UCSF CHIMERA.

Acknowledgements

This work was supported by the EU FP7 framework

program 4DCellFate. We thank I. Poser and A.

Hyman for providing the BAC GFP–CDK2AP1 cell

line. We thank N. Vermeulen-Hubner and H. Stunnen-

berg for providing access to the Q Exactive mass

spectrometer.

Author contributions

SLK, HIB, VG, PWTCJ, MB, HN, and MV per-

formed experiments. SLK, HIB, MM, and VG ana-

lyzed data. GJK contributed reagents. SLK, MM and

MV wrote the paper with input from all authors.

References

1 Zhang Y, LeRoy G, Seelig HP, Lane WS & Reinberg

D (1998) The dermatomyositis-specific autoantigen Mi2

is a component of a complex containing histone

deacetylase and nucleosome remodeling activities. Cell

95, 279–289.

2 Wade PA, Jones PL, Vermaak D & Wolffe AP (1998)

A multiple subunit Mi-2 histone deacetylase from

Xenopus laevis cofractionates with an associated Snf2

superfamily ATPase. Curr Biol 8, 843–846.

3 Tong JK, Hassig CA, Schnitzler GR, Kingston RE &

Schreiber SL (1998) Chromatin deacetylation by an

ATP-dependent nucleosome remodelling complex.

Nature 395, 917–921.

4 Spruijt CG, Bartels SJ, Brinkman AB, Tjeertes JV,

Poser I, Stunnenberg HG & Vermeulen M (2010)

CDK2AP1/DOC-1 is a bona fide subunit of the Mi-2/

NuRD complex. Mol BioSyst 6, 1700–1706.

5 Allen HF, Wade PA & Kutateladze TG (2013) The

NuRD architecture. Cell Mol Life Sci 70, 3513–3524.

10 FEBS Journal (2014) ª 2014 FEBS

Dynamics and architecture of the NuRD complex S. L. Kloet et al.

6 Miccio A, Wang Y, Hong W, Gregory GD, Wang H,

Yu X, Choi JK, Shelat S, Tong W, Poncz M, et al.

(2010) NuRD mediates activating and repressive

functions of GATA-1 and FOG-1 during blood

development. EMBO J 29, 442–456.

7 Lauberth SM & Rauchman M (2006) A conserved 12-

amino acid motif in Sall1 recruits the nucleosome

remodeling and deacetylase corepressor complex. J Biol

Chem 281, 23922–23931.

8 Eberl HC, Spruijt CG, Kelstrup CD, Vermeulen M &

Mann M (2013) A map of general and specialized

chromatin readers in mouse tissues generated by label-

free interaction proteomics. Mol Cell 49, 368–378.

9 Baubec T, Ivanek R, Lienert F & Schubeler D (2013)

Methylation-dependent and -independent genomic

targeting principles of the MBD protein family. Cell

153, 480–492.

10 Whyte WA, Bilodeau S, Orlando DA, Hoke HA,

Frampton GM, Foster CT, Cowley SM & Young RA

(2012) Enhancer decommissioning by LSD1 during

embryonic stem cell differentiation. Nature 482, 221–

225.

11 Reynolds N, Latos P, Hynes-Allen A, Loos R, Leaford

D, O’Shaughnessy A, Mosaku O, Signolet J, Brennecke

P, Kalkan T, et al. (2012) NuRD suppresses

pluripotency gene expression to promote transcriptional

heterogeneity and lineage commitment. Cell Stem Cell

10, 583–594.

12 Rais Y, Zviran A, Geula S, Gafni O, Chomsky E,

Viukov S, Mansour AA, Caspi I, Krupalnik V, Zerbib

M, et al. (2013) Deterministic direct reprogramming of

somatic cells to pluripotency. Nature 502, 65–70.

13 Dos Santos RL, Tosti L, Radzisheuskaya A, Caballero

IM, Kaji K, Hendrich B & Silva JC (2014) MBD3/

NuRD facilitates induction of pluripotency in a

context-dependent manner. Cell Stem Cell 15, 102–

110.

14 Smits AH, Jansen PW, Poser I, Hyman AA &

Vermeulen M (2013) Stoichiometry of chromatin-

associated protein complexes revealed by label-free

quantitative mass spectrometry-based proteomics.

Nucleic Acids Res 41, e28.

15 Mousson F, Kolkman A, Pijnappel WW, Timmers HT

& Heck AJ (2008) Quantitative proteomics reveals

regulation of dynamic components within TATA-

binding protein (TBP) transcription complexes. Mol

Cell Proteomics 7, 845–852.

16 Walzthoeni T, Leitner A, Stengel F & Aebersold R

(2013) Mass spectrometry supported determination of

protein complex structure. Curr Opin Struct Biol 23,

252–260.

17 van den Berg DL, Snoek T, Mullin NP, Yates A,

Bezstarosti K, Demmers J, Chambers I & Poot RA

(2010) An Oct4-centered protein interaction network in

embryonic stem cells. Cell Stem Cell 6, 369–381.

18 Wang Y, Zhang H, Chen Y, Sun Y, Yang F, Yu W,

Liang J, Sun L, Yang X, Shi L, et al. (2009) LSD1 is a

subunit of the NuRD complex and targets the

metastasis programs in breast cancer. Cell 138, 660–

672.

19 Schwanhausser B, Busse D, Li N, Dittmar G,

Schuchhardt J, Wolf J, Chen W & Selbach M (2011)

Global quantification of mammalian gene expression

control. Nature 473, 337–342.

20 Alqarni SS, Murthy A, Zhang W, Przewloka MR, Silva

AP, Watson AA, Lejon S, Pei XY, Smits AH, Kloet

SL, et al. (2014) Insight into the architecture of the

NuRD complex: structure of the RbAp48–MTA1 sub-

complex. J Biol Chem 289, 21844–21855.

21 Kuzmichev A, Nishioka K, Erdjument-Bromage H,

Tempst P & Reinberg D (2002) Histone

methyltransferase activity associated with a human

multiprotein complex containing the Enhancer of Zeste

protein. Genes Dev 16, 2893–2905.

22 Zhang Y, Iratni R, Erdjument-Bromage H, Tempst P &

Reinberg D (1997) Histone deacetylases and SAP18, a

novel polypeptide, are components of a human Sin3

complex. Cell 89, 357–364.

23 Yang B, Wu YJ, Zhu M, Fan SB, Lin J, Zhang K, Li

S, Chi H, Li YX, Chen HF, et al. (2012) Identification

of cross-linked peptides from complex samples. Nat

Methods 9, 904–906.

24 Soding J, Biegert A & Lupas AN (2005) The HHpred

interactive server for protein homology detection and

structure prediction. Nucleic Acids Res 33, W244–W248.

25 Sali A & Blundell TL (1993) Comparative protein

modelling by satisfaction of spatial restraints. J Mol

Biol 234, 779–815.

26 Eswar N, Webb B, Marti-Renom MA, Madhusudhan

MS, Eramian D, Shen MY, Pieper U & Sali A (2007)

Comparative protein structure modeling using

MODELLER. Curr Protoc Protein Sci, Chapter 2,

Unit 2.9.

27 Hauk G, McKnight JN, Nodelman IM & Bowman GD

(2010) The chromodomains of the Chd1 chromatin

remodeler regulate DNA access to the ATPase motor.

Mol Cell 39, 711–723.

28 Kahraman A, Malmstrom L & Aebersold R (2011)

Xwalk: computing and visualizing distances in

cross-linking experiments. Bioinformatics 27,

2163–2164.

29 Millard CJ, Watson PJ, Celardo I, Gordiyenko Y,

Cowley SM, Robinson CV, Fairall L & Schwabe JW

(2013) Class I HDACs share a common mechanism

of regulation by inositol phosphates. Mol Cell 51,

57–67.

30 Gnanapragasam MN, Scarsdale JN, Amaya ML, Webb

HD, Desai MA, Walavalkar NM, Wang SZ, Zu Zhu S,

Ginder GD & Williams DC Jr (2011) p66Alpha-MBD2

coiled-coil interaction and recruitment of Mi-2 are

11FEBS Journal (2014) ª 2014 FEBS

S. L. Kloet et al. Dynamics and architecture of the NuRD complex

critical for globin gene silencing by the MBD2–NuRD

complex. Proc Natl Acad Sci USA 108, 7487–7492.

31 Goujon M, McWilliam H, Li W, Valentin F, Squizzato

S, Paern J & Lopez R (2010) A new bioinformatics

analysis tools framework at EMBL-EBI. Nucleic Acids

Res 38, W695–W699.

32 Smits AH & Vermeulen M (2014) Exploring chromatin

readers using high-accuracy quantitative mass

spectrometry-based proteomics. In Systems Analysis of

Chromatin-related Protein Complexes in Cancer (Emili

A, Greenblatt J & Wodak S, eds), pp. 133–148.

Springer, New York.

33 Le Guezennec X, Vermeulen M, Brinkman AB,

Hoeijmakers WA, Cohen A, Lasonder E &

Stunnenberg HG (2006) MBD2/NuRD and MBD3/

NuRD, two distinct complexes with different

biochemical and functional properties. Mol Cell Biol 26,

843–851.

34 Malovannaya A, Lanz RB, Jung SY, Bulynko Y,

Le NT, Chan DW, Ding C, Shi Y, Yucer N,

Krenciute G, et al. (2011) Analysis of the human

endogenous coregulator complexome. Cell 145, 787–799.

35 Baymaz HI, Fournier A, Laget S, Ji Z, Jansen PW,

Smits AH, Ferry L, Mensinga A, Poser I, Sharrocks A,

et al. (2014) MBD5 and MBD6 interact with the

human PR–DUB complex through their methyl-CpG-

binding domain. Proteomics doi: 10.1002/

pmic.201400013. [Advance online publication].

36 Dignam JD, Lebovitz RM & Roeder RG (1983)

Accurate transcription initiation by RNA polymerase II

in a soluble extract from isolated mammalian nuclei.

Nucleic Acids Res 11, 1475–1489.

37 Kessner D, Chambers M, Burke R, Agus D & Mallick

P (2008) ProteoWizard: open source software for rapid

proteomics tools development. Bioinformatics 24, 2534–

2536.

38 Wang LH, Li DQ, Fu Y, Wang HP, Zhang JF, Yuan

ZF, Sun RX, Zeng R, He SM & Gao W (2007)

pFind 2.0: a software package for peptide and protein

identification via tandem mass spectrometry. Rapid

Commun Mass Spectrom 21, 2985–2991.

39 Pettersen EF, Goddard TD, Huang CC, Couch GS,

Greenblatt DM, Meng EC & Ferrin TE (2004) UCSF

Chimera – a visualization system for exploratory

research and analysis. J Comput Chem 25, 1605–1612.

Supporting information

Additional supporting information may be found in

the online version of this article at the publisher’s web

site:Table S1. iBAQ values used to calculate NuRD sub-

unit stoichiometry.

Table S2. Overview of all assigned cross-links with

FDR < 0.05 as calculated in PLINK.

12 FEBS Journal (2014) ª 2014 FEBS

Dynamics and architecture of the NuRD complex S. L. Kloet et al.