nature biotechnology  VOLUME 29 NUMBER 3 MARCH 2011 255

A rt i c l e s

Protein lysine acetylation is a key mechanism in the epigenetic con-trol of gene expression and the regulation of cell metabolism1–3, and protein deacetylases are potential targets for treating cancer and a range of autoimmune and neurodegenerative diseases4. The first mammalian histone deacetylase (HDAC) was discovered in 1996 by a chemical biology approach using an immobilized, microbially derived compound as affinity matrix5.

Based on sequence phylogeny and function, there are four distinct classes of HDAC: class I (HDAC1, 2, 3 and 8), class IIa (HDAC4, 5, 7 and 9), class IIb (HDAC6 and 10) and class IV (HDAC11) represent Zn2+-dependent amidohydrolases, whereas class III comprises the mechanistically diverse NAD+-dependent sirtuins6. HDACs form the catalytic core of megadalton complexes involved in chromatin modification and gene repression. Four such molecular machines have been characterized to date. Whereas the CoREST, NuRD and Sin3 complexes contain an HDAC1-HDAC2 dimer as core, the NCoR complex is formed around HDAC3 (ref. 7). The roles of these com-plexes are diverse and often cell-type specific. Although more data are emerging regarding their role in the determination of cell fate, their functions in tissue homeostasis are less well understood8. The CoREST complex couples histone deacetylation to demethylation to repress neuronal genes9, the NuRD complex links deacetylation to a chromatin-remodeling ATPase and promotes gene silencing10, and the Sin3 complex represses genes downstream of various devel-opmental pathways11. The NCoR complex is the major corepressor for nuclear receptors12,13. Class IIa HDACs exhibit low enzymatic

activity and are proposed to have “modification reader” or scaffold functions14,15. Class IIb HDACs exhibit mostly nonepigenetic functions in regulating protein folding and turnover16.

Small-molecule HDAC inhibitors were discovered by their ability to induce redifferentiation of transformed cells17. Suberoylanilide hydroxamic acid (SAHA; vorinostat, Zolinza) and romidepsin (Istodax) are approved for the treatment of cutaneous T-cell lym-phoma, and valproate is in clinical use as an anticonvulsant. Several HDAC inhibitors are in development for a number of indications but clinical development has been hampered by a lack of target selectiv-ity. This increases the risk of toxic liabilities and also limits the use of these compounds as research tools18.

The perceived lack of selectivity of HDAC inhibitors may originate from the optimization of lead compounds using standard industry assays involving recombinant enzymes or protein fragments. These seem unlikely to properly reflect the native conformation and activity of the target and its physiological context owing to incorrect protein folding, post-translational modifications and the absence of regulatory subunits. Remarkably, purified class I HDACs exhibit increased activ-ity in the presence of interacting proteins13,19. Most HDAC inhibitors adhere to a distinctive pharmacophore comprising a ‘cap’, which binds to the rim of the substrate channel, a spacer spanning the channel, and a Zn2+-chelating function. A photoaffinity analog of SAHA was shown to label not only HDACs but also the proteins RCOR1, MBD3 and MTA1/2. This indicates that these proteins are close to the active site, and suggests that the cap conveys inhibitor selectivity20.

chemoproteomics profiling of HDAc inhibitors reveals selective targeting of HDAc complexesMarcus Bantscheff1,3, Carsten Hopf1,3, Mikhail M Savitski1, Antje Dittmann1, Paola Grandi1, Anne-Marie Michon1, Judith Schlegl1, Yann Abraham1, Isabelle Becher1, Giovanna Bergamini1, Markus Boesche1, Manja Delling1, Birgit Dümpelfeld1, Dirk Eberhard1, Carola Huthmacher1, Toby Mathieson1, Daniel Poeckel1, Valérie Reader2, Katja Strunk1, Gavain Sweetman1, Ulrich Kruse1, Gitte Neubauer1, Nigel G Ramsden2 & Gerard Drewes1

The development of selective histone deacetylase (HDAC) inhibitors with anti-cancer and anti-inflammatory properties remains challenging in large part owing to the difficulty of probing the interaction of small molecules with megadalton protein complexes. A combination of affinity capture and quantitative mass spectrometry revealed the selectivity with which 16 HDAC inhibitors target multiple HDAC complexes scaffolded by ELM-SANT domain subunits, including a novel mitotic deacetylase complex (MiDAC). Inhibitors clustered according to their target profiles with stronger binding of aminobenzamides to the HDAC NCoR complex than to the HDAC Sin3 complex. We identified several non-HDAC targets for hydroxamate inhibitors. HDAC inhibitors with distinct profiles have correspondingly different effects on downstream targets. We also identified the anti-inflammatory drug bufexamac as a class IIb (HDAC6, HDAC10) HDAC inhibitor. Our approach enables the discovery of novel targets and inhibitors and suggests that the selectivity of HDAC inhibitors should be evaluated in the context of HDAC complexes and not purified catalytic subunits.

1Cellzome AG, Heidelberg, Germany. 2Cellzome Ltd., Chesterford Research Park, Cambridge, United Kingdom. 3These authors contributed equally to this work. Correspondence should be addressed to M.B. (marcus.bantscheff@cellzome.com) or G.D. (gerard.drewes@cellzome.com).

Received 18 November 2010; accepted 17 December 2010; published online 23 January 2011; doi:10.1038/nbt.1759

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

256  VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

A rt i c l e s

Recent advances in chemoproteomics enabled binding studies of small-molecule enzyme inhibitors to endogenous proteins in cells and tissues21–23. Here we extend the chemoproteomics approach from the monitoring of individual target proteins to the analysis of inhibitors bind-ing to native megadalton protein complexes, with a view to discovering novel targets, complexes and inhibitors. We found that HDAC inhibitors targeted known and novel protein complexes that are precisely defined by matching half-maximal inhibitory concentration (IC50) values for a given inhibitor for all complex subunits, and we used quantitative immuno-affinity purifications to confirm the composition of the complexes. Inhibitor selectivity data for native drug target complexes deviated from literature values obtained using isolated recombinant enzymes, indicating an unexpected degree of selectivity of certain HDAC inhibitors.

RESULTSSynthesis of a target class–specific HDAC probe matrixTarget class–directed chemical probes provide tools for the identifi-cation of drug targets directly in cells and tissues. Typically, probes

consist of a moiety that binds to a ligand pocket conserved within the target class under investigation, and a functional group for immobili-zation enabling efficient enrichment of bound proteins for analysis24. HDACs share a conserved substrate pocket, and most hydroxamate inhibitors are nonselective25.

We synthesized a target class–specific probe matrix by derivatiz-ing sepharose with analogs of the hydroxamates SAHA and givino-stat (ITF2357). The probe matrix was exposed to cell extracts, and aliquots of the sample were treated with excess inhibitor, which competes with the immobilized probes for target protein binding. The reduction in protein capture that resulted from inhibitor treat-ment was quantified by isobaric tagging of tryptic peptides and tan-dem mass spectrometry analysis (MS/MS) of the combined peptide pools26. For each identified protein, the decrease of the reporter ion signals relative to the vehicle control reflects the competitive binding of the ‘free’ inhibitor to its target. The results comprise binding data for both direct enzyme targets and proteins residing in a complex with the target, as these are predicted to have matching IC50 profiles

1

2 3 4 5 6

Vehiclecontrol

TMT 131

TMT 130

TMT 129

TMT 128

TMT 127

TMT 126

Inhi

bito

r co

ncen

trat

ion Target binds to drug

131 m/z

130

129

128

127

126

0

100

50

Protein does not bind to drug

131 m/z

130

129

128

127

126

0

100

50

131 m/z

130

129

128

127

126

0

100

50

Protein C binds to target

131 m/z

130

129

128

127

126

0

100

50

Protein D binds to target

Probe matrixa

b

log [inhibitor]

Target

log [inhibitor]

log [inhibitor]

Protein CProtein D

131 m/z

130

129

128

127

126

0

100

50

Protein A binds to target

131 m/z

130

129

128

127

126

0

100

50

Protein B binds to target

log [inhibitor]

log [inhibitor]

Protein AProtein B

IP of protein A

IP of protein D

HN

O

NH

O

OH

O

NH

NH

O

O

OH

Figure 1 Mapping of HDAC drug target complexes in chemical space and in proteome space. (a) Chemoproteomics competition binding assay to profile HDAC inhibitor target complexes in cell extract. (1) A probe matrix is generated by derivatizing sepharose with analogs of nonselective HDAC inhibitors (left, SAHA, right, givinostat). (2) Cell extract is incubated with vehicle or with drug over a range of concentrations. (3) The ‘free’ drug competes with the immobilized probes for drug-binding sites on target-protein complexes. White hexagon, inhibitor drug. (4) Captured proteins are trypsinized and each peptide mixture is tagged with a distinct isobaric tandem mass tag (TMT). (5) Tagged samples are pooled and analyzed by LC-MS/MS. Each peptide gives rise to six reporter signals in the MS/MS spectrum. (6) When free drug outcompetes protein capture, signal intensities relative to the vehicle control decrease for each peptide originating from this protein. Complexes formed by the target and associated proteins are defined by matching inhibition (IC50) curves. (b) Definition of target protein complexes in biological space by quantitative co-IP. Data generated from the same cell extracts are used to deconvolute protein complexes formed around the drug target.

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnology  VOLUME 29 NUMBER 3 MARCH 2011 257

A rt i c l e s

across a set of drugs. We confirmed associations of proteins in com-plexes using immunoaffinity purifications (Fig. 1).

Quantitative mapping of protein binding to the probe matrixWe first tested the probe matrix with recombinant HDACs 1–11 purified from Sf9 insect cells. With the exception of HDAC1, the enzymes were found to bind to the matrix. However, binding was only partially reduced by excess SAHA, indicating that the bulk of the purified enzyme exhibited low activity (Supplementary Table 1). HDAC3, the only enzyme purified together with a cofactor (NCoR2), was most susceptible to competition with excess SAHA. The activ-ity of the enzymes in a peptide deacetylation assay (Supplementary Fig. 1) may arise from a small fraction of properly folded protein or from contamination with insect-cell activities15,27. Consistent with the binding data, HDAC1 showed the lowest activity, whereas HDAC3, used as a complex with NCoR2, was most active.

In contrast to the recombinant proteins, all class I and class IIb HDACs and many known HDAC complex subunits were specifically captured when the SAHA- or givinostat-derivatized matrix was used to probe endogenous HDACs in cell extracts from the myelogenous leukemia line K562 (Supplementary Fig. 2 and Supplementary Data Set 1). The ability to competitively inhibit the binding of a protein to the probe matrix with excess ‘free’ inhibitor distinguishes specific binding from nonspecific background. From the ~2,600 pro-teins identified, 267 proteins exhibited substantially reduced matrix

binding when excess SAHA or trichostatin A (TSA) was added to the lysate, with their corresponding reporter ion intensities reduced by 50% or more. These were designated as potential inhibitor targets, and included the six class I and class IIb HDACs, and 29 proteins known to associate with class I HDACs in the CoREST, NuRD, Sin3 and NCoR complexes. In addition, many proteins previously not associated with deacetylase function bound to the probe matrix specifically, implicat-ing them either as novel inhibitor targets or as components of target complexes. For instance, 5,10-methenyltetrahydrofolate synthetase and hydroxysteroid (17-beta) dehydrogenase 4 were specifically cap-tured by givinostat-conjugated, but not SAHA-conjugated, matrix. This suggests that they are targets of givinostat but not SAHA.

Next, the subproteome binding to the SAHA matrix was differen-tially mapped in nuclear versus cytosolic fractions of Jurkat cells, and in a panel of human cell lines and mouse tissues (Supplementary Fig. 2 and Supplementary Data Set 1). Most proteins in the subproteome appeared to be ubiquitously expressed. As expected, the majority was enriched in the nuclear fraction. Because of the function of class I HDACs in cell division28, we conducted differ-ential mapping of the subproteome in HeLa cells arrested in mito-sis or G1/S phase, compared to nonarrested cells. Notably, three proteins were specifically captured by the probe matrix in greater amounts from mitotic cells compared to nonmitotic cells, and thus may constitute a novel HDAC complex. These are DNTTIP1 (deoxynucleotidyltransferase-interacting protein), C14ORF43,

BML210

NH

NH2

NH2

O

O

HN

Amino-benzamide

Res

idua

l bin

ding

0

1.0

1 10 100 1,000 1 10 100 1,000 1 10 100 1,000

Tacedinaline (CI-994)

NH

O NH

O

Res

idua

l bin

ding

0

1.0

1 10 100 1,000 1 10 100 1,000 1 10 100 1,000

[Compound] (µM) [Compound] (µM) [Compound] (µM)

Romidepsin

Res

idua

l bin

ding

0

1.0

0.001 0.01 0.1 1

HDAC1HDAC2HDAC3HDAC6HDAC8HDAC10

0.001 0.01 0.1 1

GATAD2ALSD1MTA3NCOR1RCOR1SIN3ATBL1XR1

0.001 0.01 0.1 1

BZW2DNTTIP1ISOC2MIDEASPDXK

HDACs Complex subunits Other proteins

SAHA

HN

O

NH

O

OH

Hydroxamicacid R

esid

ual b

indi

ng

0

1.0

0.01 0.1 1 10 0.01 0.1 1 10 0.01 0.1 1 10

a

0

Rel

ativ

e po

tenc

y

1

CDK2AP1CHD4MBD2EHMT2REREZMYM3MIER1RREB1MIDEASCDYLDNTTIP1ZMYM2MTA2RBBP4GATAD2AGATAD2BPHF21AMIER2MBD3MTA1MTA3

ISOC2LOC153364HDAC6BRMS1WIZSIN3BSAP30LSIN3ASAP30

MIER3HDAC2HDAC1RCOR2HMG20BRCOR3GSE1LSD1HMG20ARCOR1

99

HDAC3GPS2NCOR1

NC

OR

1S

IN3

CoR

ES

T

NCOR2TBL1XTBL1XR1

100

98

95

PCI−34

051

MC−1

293

Valpro

ic ac

id

PCI−24

781

Romide

psin

Taced

inalin

e

Entino

stat

BML−

210

Moc

etino

stat

Script

aid

SAHA

Belino

stat

Apicidi

n

Panob

inosta

t

Dacino

stat

Tricho

statin

A

b

NH

O

NH

O

NH

ONH

O

O

O

S

S

Figure 2 HDAC inhibitor drug targets and target complexes are defined by chemoproteomics profiling of drugs and compounds used as research tools. (a) Representative concentration-inhibition profiles of SAHA, BML-210 (aminobenzamide analog of SAHA), tacedinaline and romidepsin were determined in K562 cell extract as outlined in Figure 1. Inhibitors were pre-incubated with cell extracts at 4 °C before addition of the probe matrix, with the exception of the aminobenzamides BML-210 and tacedinaline, which were pre-incubated at 22 °C. Profiles are grouped in three plots for each inhibitor: HDACs (left), components of CoREST, NuRD, Sin3 and NCoR complexes (middle), and examples of other proteins either representing novel direct targets or complex components (right). Previously known complex associations are represented in a color code. Profiles of additional inhibitors are depicted in Supplementary Figure 3. (b) Bidirectional hierarchical clustering of the concentration-inhibition data for 16 inhibitors versus 1,251 proteins (each targeted by at least one inhibitor). Only the area of the clustering around HDACs is shown. For better comparison of selectivities, average pKd

app values were transformed into relative affinities scaling from 0 to 1 for each inhibitor. Statistically significant clusters are highlighted in blue and brown representing >95% and >99% unbiased bootstrap probability, respectively (Supplementary Fig. 7).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

258  VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

A rt i c l e s

a protein of unknown function, which we dubbed MIDEAS (for mitotic deacetylase-associated SANT domain protein), and the putative histone acetylase CDYL.

Proteomics target profiling of HDAC inhibitors The probe matrix offers a unique tool to probe a subproteome of putative targets with drugs under close to physiological conditions. We selected a set of 16 structurally diverse HDAC inhibitors, includ-ing approved drugs, compounds in clinical development and small molecules with potential as research tools (Supplementary Table 2). Aliquots of K562 cell extract were incubated with vehicle or different

inhibitors at five concentrations, typically ranging from 40 nM to 10 µM. For low- or very high-potency compounds, concentrations were adjusted accordingly. Subsequently, samples were incubated with the probe matrix, captured proteins were quantitatively mapped by MS/MS and a set of IC50 values was determined for each inhibitor (Fig. 2a and Supplementary Fig. 3). To assess reproducibility, we carried out several replicate experiments per inhibitor profile using targeted data acquisition29 (Supplementary Data Set 2).

The data were sufficiently reproducible to discriminate fairly small (twofold) differences in IC50 values. Remarkably, we found statisti-cally significant differences between distinct complexes containing

Table 1 Kdapp values (in mM) for selected HDAC inhibitors of molecular targets and target complexes, as determined using chemoproteomic

binding profilingBelinostat

preincub. at 4 °C max conc. 30 µM

BML-210 preincub. at 22 °C max conc. 300 µM

Romidepsin preincub. at 4 °C

max conc. 100 nM

SAHA preincub. at 4 °C max conc.

10 µM

Tacedinaline preincub. at 22 °C max conc. 600 µM

Trichostatin A preincub. at 4 °C max conc. 1 µM

Valproic acid preincub. at 4 °C max conc. 20 mM

Classification Protein Kdappa s.d.b Kd

app s.d. Kdapp s.d. Kd

app s.d. Kdapp s.d. Kd

app s.d. Kdapp s.d.

Class I HDAC HDAC1 0.29 0.02 14.9 1.0 0.008 0.001 0.28 0.05 13.0 1.8 0.010 0.001 745 147HDAC2 0.35 0.04 10.7 0.2 0.010 0.001 0.35 0.08 11.9 1.4 0.012 0.002 975 249HDAC3 0.26 0.08 2.8 0.4 0.017 0.001 0.29 0.06 6.0 0.3 0.041 0.007 9,350 3949HDAC8 >30c – > 300 – > 0.1 – > 10 – 240 137 > 1 – 12,675 734

Class II HDAC HDAC6 0.53 0.14 > 300 – > 0.1 – 0.13 0.04 > 600 – 0.109 0.025 >20,000 –HDAC10 >30 – > 300 – > 0.1 – 5.26 2.34 > 600 – > 1 – >20,000 –

CoREST complex

GSE1 LSD1

0.29 0.29

0.09 0.07

9.2 9.2

0.9 0.5

0.006 0.006

0.001 0.001

0.22 0.17

0.06 0.05

15.5 16.9

0.7 1.2

0.008 0.007

0.002 0.001

594 568

158 174

HMG20A 0.31 0.09 9.8 0.4 0.005 0.001 0.22 0.04 14.7 1.2 0.006 0.001 493 95HMG20B 0.33 0.10 12.6 0.5 0.006 0.001 0.25 0.04 18.4 1.8 0.008 0.001 748 202PHF21A 0.25 0.04 31.8 1.0 0.012 – 0.27 0.05 68.6 5.6 0.013 – 509 –RCOR1 0.20 0.03 8.5 0.5 0.005 0.001 0.24 0.06 15.0 1.6 0.007 0.001 516 130RCOR2 0.21 0.06 18.2 5.5 0.008 0.000 0.13 0.02 18.0 2.2 0.008 0.001 536 172RCOR3 0.26 0.08 10.5 0.3 0.005 0.001 0.21 0.05 20.6 1.7 0.007 0.001 534 123ZMYM2 0.36 0.06 43.3 32.2 0.013 0.005 0.22 0.07 112 73 0.012 0.002 1251 327ZMYM3 0.39 0.24 134 90.9 0.035 0.010 0.58 161 32 0.015 8,23

NuRD complex CDK2AP1 2.63 59.5 – – – – – – 45.9 2.5 0.020CHD4 – – 64.1 – 0.024 0.011 0.41 – – – 0.057 2,897 8641GATAD2A 0.56 0.09 42.9 8.8 0.016 0.001 0.38 0.09 79.6 3.1 0.020 0.002 1,219 231GATAD2B 0.62 0.11 33.0 1.7 0.014 0.001 0.41 0.10 53.6 4.3 0.032 0.023 1,429 212MBD2 41.9 0.66 81.5 0.023 1,153MBD3 0.68 0.09 38.2 2.9 0.013 0.001 0.33 0.04 46.9 3.9 0.020 0.002 1,483 249MTA1 0.53 0.11 0.011 0.004 0.28 0.11 45.5 6.8 0.021 1,223 618MTA2 0.60 0.23 41.6 3.1 0.010 0.001 0.56 0.26 119 34 0.028 0.004 1,048 140MTA3 0.41 0.07 24.9 4.1 0.012 0.002 0.28 0.04 48.3 7.9 0.019 0.002 1,099 180

Sin3 complex BRMS1 0.41d – > 300 – 0.043 – – – > 600 – 0.022 – >20,000 –SAP18 – – > 300 – – – – – – – – –SAP30 0.09 0.01 > 300 – 0.029 0.002 0.17 0.04 > 600 – 0.025 0.003 10,485 3,911SAP30L 0.20 > 300 – – 0.17 – 0.049 0.015 13,236 4,056SIN3A 0.10 0.01 > 300 – 0.036 0.009 0.14 0.03 > 600 – 0.025 0.006 12,671 4,281SIN3B 0.16 0.05 > 300 – 0.027 0.016 0.26 0.14 > 600 – 0.024 15,690 2,739

NCoR complex GPS2 0.31 0.11 2.9 0.1 0.016 0.001 0.20 0.07 6.6 0.5 0.027 0.005 6,068 4,825NCOR1 0.25 0.07 1.8 – 0.017 0.002 0.26 0.06 5.7 0.7 0.024 0.000 3,598 365NCOR2 0.35 0.02 1.9 0.1 0.017 0.002 0.21 0.06 6.8 1.2 0.026 0.001 4614 2,471TBL1X 0.29 0.13 2.6 0.1 0.018 0.002 0.26 0.05 4.9 0.3 0.029 0.002 4,673 1,160TBL1XR1 0.27 0.09 2.3 0.1 0.014 0.001 0.27 0.04 4.9 0.3 0.025 0.003 2,956 951

DNTTIP1 DNTTIP1 0.22 0.07 55.5 4.4 0.014 0.001 0.18 0.03 119 20 0.014 0.002 1,187 286MIDEAS 0.31 0.05 162 78.2 0.013 0.002 0.18 0.09 156 28 0.013 0.003 1,110 371

ELM-SANT domain proteins

MIER1 MIER2

0.24 0.59

0.03 0.22

21.4 26.3

3.7 –

0.023 0.003 –

0.25 –

0.05 36.1 37.0

3.9 –

0.016 –

0.002 –

1,423 –

892 –

MIER3 0.11 0.01 6.3 – 0.009 0.001 0.12 0.04 – – 0.007 – 954 181RERE 0.49 0.12 222 48.2 0.039 0.005 0.32 0.20 272 57 0.025 0.009 4,336 1,131

Selected HDACi off-targets

ALDH1A2 BZW2

> 30 2.14

– 0.26

> 300 > 300

– –

> 0.1 > 0.1

– –

0.66 2.72

0.06 0.40

> 600 > 600

– –

> 1 0.96

– 0.02

10,528 >20,000

5,352 –

CBR1 1.90 0.07 > 300 – > 0.1 – > 10 – > 600 – > 1 – >20,000 –ISOC2 > 30 – > 300 – > 0.1 – 0.06 0.01 > 600 – > 1 – >20,000 –PPP3CA 3.90 0.57 > 300 – > 0.1 – > 10 – > 600 – > 1 – >20,000 –

aNo Kdapp listed: protein was not detected in sample. bNo s.d. listed: protein was only detected in one sample or not at all. c >, no inhibition at maximum compound concentration tested (in at

least two experiments). ditalics, IC50 value listed (no Kdapp was determined).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnology  VOLUME 29 NUMBER 3 MARCH 2011 259

A rt i c l e s

HDAC1 and HDAC2 (Supplementary Fig. 4 and Supplementary Data Set 3). The quanti-fication of relative protein amounts sequestered by the probe matrix upon sequential incubations enabled us to determine apparent dissociation constants (Kd

app) from the IC50 values. The resulting deviation between Kd

app and IC50 values was less than twofold for 99% of the proteins (Supplementary Data Set 4). It is interesting to compare the target profiles of SAHA and its analog BML-210, which are identical except for the replacement of the hydroxamate by an aminobenzamide group in BML-210. This change causes a general drop of potency for class I HDACs (and complete loss of potency for class IIb), concomitant with an increase in selectivity for HDAC3 relative to HDAC1 or HDAC2 (Fig. 2a). Some compounds of the aminobenzamide class were reported to exhibit slow binding kinetics for class I HDACs, in particular for an HDAC3-NCoR2 complex30. We explored the effect of preincubation of aminobenzamide inhibitors with the cell extract at 22 °C, or pro-longed preincubation at 4 °C. Under either condition we observed a more pronounced inhibition of HDAC3 binding to the probe matrix (Supplementary Fig. 5). Consequently, we performed additional

profiling experiments for all aminobenzamides using preincubation at 22 °C and found that inhibition of all the class I HDACs was affected, with the greatest change in inhibitor potency for HDAC1 and HDAC2 (up to fivefold for some compounds; Supplementary Fig. 3).

The data set comprises concentration-inhibition profiles for 16 compounds, assayed with 1,251 proteins. Combinations with at least twofold reduction in binding by any of the inhibitors are shown (Fig. 2b). The set of proteins included the class I and IIb HDACs, components of HDAC complexes, and putative novel complex com-ponents or targets (data for selected inhibitors listed in Table 1, additional inhibitors in Supplementary Table 3 and MS data in Supplementary Data Set 2). Because the profiles were generated using high maximum inhibitor concentrations, it is unlikely that all of these proteins are physiologically relevant targets. The inhibition curves typically displayed a Hill coefficient of ~1 (indicating stoichio-metric binding of one inhibitor molecule per enzyme molecule), with

HDAC1Enrichment in IP vs ctrl IgG

HDAC(class I)

HDAC1HDAC2HDAC3

CoREST GSE1HMG20AHMG20BLSD1PHF21ARCOR1RCOR2RCOR3ZMYM2ZMYM3ZNF217

NCoR GPS2NCOR1NCOR2TBL1XTBL1XR1

NuRD CDK2AP1CHD3CHD4GATAD2AGATAD2BMBD2MBD3MTA1MTA2MTA3

NuRD/Sin3(shared)

RBBP4RBBP7

Sin3 ARID4AARID4BBRMS1BRMS1L*ING1*ING2*SAP30SAP30LSAP130SIN3ASIN3BSUDS3

DNTTIP1 DNTTIP1MIDEAS

ELM-SANT MIER1MIER2MIER3RERETRERF1

CDYL

Histones HIST1H1CHIST1H4AHIST2H2BA

HATs CDYLHistone methylation

EHMT1EHMT2WDR5WIZ

EHMT2

0.5 1.0ident. in IP:

Baits:

HDAC2

LSD1M

TA3

SIN3A

DNTTIP1

TRERF1

HDAC3

TBL1XR1

a

Relative enrichment in LSD1 in IP

Relative enrichment in MTA3 in IP

1 HDAC12 HDAC23 LSD1 (bait)4 HMG20A5 RCOR36 RCOR17 GSE18 SIN3A (bait)9 SAP3010 ARID4B11 SAP3012 SUDS3A13 BRMS114 RBBP415 RBBP7

1 HDAC12 HDAC23 HDAC34 TBL1XR1 (bait)5 NCOR16 TBL1X7 GPS28 MTA3 (bait)9 CDK2AP110 CHD311 CHD412 GATAD2A13 GATAD2B14 MBD215 MBD316 MTA117 = MTA2

Rel

ativ

e en

richm

ent i

n S

in3A

in IP

Rel

ativ

e en

richm

ent i

n T

BL1

XR

1 in

IP

0

0

–0.2

–0.2–0.4

–0.4

0.2

0.2

0.4

0.4

0.6

0.6

0.8

0.8

1.0

0

–0.2

–0.4

0.2

0.4

0.6

0.8

1.0

1.0

0–0.2–0.4 0.2 0.4 0.6 0.8

2

1

910

11

1314

15

128

5

3

4

7 6

2 1

5347

6

CoREST

NuRD

NCoR

SIN3

16

8

9 17

1314

1512

b

DNTTIP1 TRERF1 CDYL EHMT2

0 1Relative affinity 0 1Relative affinity 0 1Relative affinity 0 1Relative affinity

0.5

1.0

3130

2

1

30

1

41

38

37

28

2

45

4342

41

37

2

HDAC1 HDAC2 HDAC3

0 1 0 1 0 1

0.5

1.0

37

31

302928

27

26

24

23

2221

2019

18

1716

1514

111 9

76 5

4

21

45444342

4140 393837

3130

29 2827 26

2524

23

22212019

18

17

16

15

1413

12

11

1098

7

6

5

4

21

36

35

343332

13

7

3

Enr

ichm

ent I

P

LSD1 MTA3 SIN3A TBL1XR10 1 0 1 0 1 0 1

0.5

1.0

119

7

6

5

4

2

129

282221

20

1918 17

16

15

14 2

1

29

282625

24

231

3635

33

32

7

3

Enr

ichm

ent I

PE

nric

hmen

t IP

HDAC3 3 SAP30L 25

HMG20A 5 SIN3B 27

LSD1 7 RBBP7 29

RCOR1 9 MIDEAS 31

RCOR3 11 NCOR1 33ZMYM2 12 NCOR2 34ZMYM3 13 TBL1X 35

GATAD2A 16 MIER2 38GATAD2B 17 MIER3 39MBD2 18 RERE 40

MTA1 20 EHMT1 42MTA2 21 EHMT2 43MTA3 22 RREB1 44

WIZ 45

cGSE1 SIN3A 264CoREST

HDAC2 SAP30 242(class I)

CDK2AP1 TBL1XR1 3614NuRD

Protein ProteinNo.Group No. GroupHDAC1 BRMS11HDAC 23Sin3

HMG20B 6 RBBP4 28NuRD/Sin3

PHF21A 8 DNTTIP1 30DNTTIP1

RCOR2 10 GPS2 32NCoR

CHD4 15 MIER1 37ELM-SANT

MBD3 19 CDYL 41Other

Figure 3 Deconvolution of protein complexes by co-IP analysis confirms the identification of novel HDAC complexes. (a) HDAC complexes identified by both chemoproteomics profiling and co-IP–MS/MS analysis of HDAC complexes. IPs were performed from K562 cells using antibodies for HDAC1, 2 and 3, known complex components (the CoREST subunit LSD1, the NuRD subunit MTA3, the Sin3 subunit SIN3A, and the NCoR-subunit TBL1XR1) and examples of novel HDAC interacting proteins. * denotes previously reported complex components not captured by the SAHA matrix. The color code indicates enrichment E of immunoprecipitated proteins as compared to mock-IP experiments (scales from −1 to 1, E = 0 denotes equal abundance, see Online Methods). (b) Examples of the quantitative mapping of immunoaffinity-purified protein complexes by MS/MS. Purifications conducted with two different antibodies each, and corresponding isotype controls, were combined after PAGE, trypsinization and isobaric tagging. Quantification data are shown as plots of relative enrichment in immunoprecipitates of Sin3 versus LSD1 (upper panel), and MTA3 versus TBL1XR1 (lower panel). Each square represents a protein with its size scaled according to the number of sequence-to-spectrum matches. (c) HDAC protein complexes in chemical and protein space. For each protein identified in chemoproteomics and co-IP experiments, enrichment in the IP samples is plotted against the average relative affinity data across all inhibitors tested. Target proteins are represented in red (class I HDACs), blue (CoREST components), green (NCoR components), purple (NuRD components), light blue (Sin3 components), pink (MiDAC components) and yellow (ELM-SANT proteins). The square size indicates the confidence of the interaction with the immunopurified protein complex (large squares: FDR < 0.05, medium-sized squares: 0.05 < FDR < 0.15, c.f. Supplementary Figs. 8 and 9 and Supplementary Table 4).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

260  VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

A rt i c l e s

the notable exception of romidepsin, which consistently yielded a Hill coefficient of 2 (Supplementary Fig. 6). Bidirectional hierarchical clustering of the complete data set clearly outlines several HDAC complexes defined by their inhibition profiles relative to the major chemical compound classes (Fig. 2b and Supplementary Fig. 7).

In the chemical dimension, the clustering is driven by the major chemotypes with several hydroxamate subclusters that differed in their effect on class II HDACs. In agreement with published data, we found that peptidic and hydroxamate compounds are substantially more potent than aminobenzamides. A notable observation was the unexpected degree of selectivity of aminobenzamide inhibitors, which show a preference for the HDAC3-NCoR complex. Clustering in the protein dimension is driven by the association of proteins in complexes, because all subunits of a target complex exhibit Kd

app values for a given inhibitor. It should nonetheless be noted that proteins known to reside in two or more complexes (e.g., HDAC1, HDAC2, RBBP4/7 or LSD1 (ref. 7) are predicted to exhibit Kd

app values representing aggregates between the individual complexes. The data show that complex subunits remained associated with the inhibitor target proteins during the assay procedure, and that they affect the inhibitor-binding properties of the catalytic HDAC subunits. Whereas the HDAC1/2-containing CoREST and NuRD complexes showed similar inhibition profiles across the com-pound set, there were marked differences in the Sin3 inhibition pro-files, with the aminobenzamides and valproate showing a much lower potency for the Sin3 complex compared to CoREST and NuRD. The clustering delineates additional HDAC complexes formed around the ELM-SANT domain proteins MIDEAS, MIER1, MIER2, MIER3 and RERE. Notably, MIDEAS, DNTTIP1 and CDYL, which were captured in greater amounts by the probe matrix from mitotic cells compared to nonmitotic cells, were clustered in close proximity, suggesting the exist-ence of a distinct mitotic deacetylase complex (MiDAC).

Deconvolution of HDAC target complexesTo differentiate between novel direct targets and novel components of HDAC complexes, we carried out a series of co-immunoprecipita-tions (co-IP) with quantitative MS/MS analysis. We evaluated a set of 27 antibodies and ultimately selected 14 directed against three class I HDACs, four known complex protein subunits (LSD1 from CoREST, MTA3 from NuRD, SIN3A from Sin3, TBL1XR1 from NCoR) and four potential HDAC complex components identified in the chemoproteomics profiles (DNTTIP1, the ELM-SANT protein TRERF1, the histone acetylase CDYL and the histone methylase EHMT2). The co-IP data sets comprise many known complex com-ponents and new targets or components already identified in the chemoproteomics profiles (Fig. 3a and Supplementary Data Set 5). Co-IP samples from two different antibodies and corresponding control IgG samples were combined for quantitative MS/MS analy-sis, such that specifically co-immunopurified proteins were clearly discriminated from background and differences in complex compo-sitions were directly quantified. Notably, the display of the SIN3A- versus LSD1-precipitated proteins indicated a preference of the Sin3 complex for HDAC1 relative to HDAC2 (Fig. 3b). However, most co-IP samples contained large numbers of co-purifying proteins, thus obscuring the identity of true interactors. For instance, ~300 proteins were at least twofold enriched with two different HDAC2 antibodies relative to control IgG samples. Evidently, HDAC inhibi-tor target complexes contained in the co-IP data sets should also be present in the chemoproteomics data (Fig. 2b). Hence we analyzed the overlap between these orthogonal data sets and devised a con-fidence score to statistically assess the association of proteins with the target protein complexes (Fig. 3c, Supplementary Fig. 8 and

Supplementary Table 4). For each protein, the relative enrichment in each co-IP and the average relative potency values across all inhib-itors were normalized to a scale from 0 to 1, and the confidence score was defined as the sum of squares of both values, scaling between 0 and 2. The calculation of this score for each protein in the experi-mental and a corresponding randomized data set enabled deter-mination of high-confidence complex associations with very low false-discovery rates (Supplementary Fig. 9).

The ELM-SANT domain proteins MIER1, MIER2, MIER3 and RERE were co-purified with HDAC1 and/or HDAC2 but not with the CoREST, NuRD, Sin3 or NCoR complexes. This is consistent with their inhibition profiles not matching any of the known complexes (Fig. 2b). These pro-teins therefore likely represent components of distinct HDAC complexes formed around ELM-SANT scaffolds31. MIER1, MIER2 and HDAC2 (but not HDAC1) were co-purified with the putative histone acetylase CDYL. MIER1, HDAC2 and CDYL were also found in the EHMT2 co-IP, suggesting that the reported CDYL-EHMT2 complex32 contains MIER1, MIER2 and HDAC2. The MiDAC complex was confirmed in the DNTTIP1 co-IP, which comprised HDAC1, HDAC2, MIDEAS and TRERF1, an ELM-SANT protein related to MIDEAS, which also co-purified with HDAC2. Immunoaffinity purification of TRERF1 itself confirmed its association with DNTTIP1 and HDAC1/2, but not with MIDEAS. This suggests that TRERF1 and MIDEAS represent alternative scaffolds for related complexes.

To further characterize the MiDAC complex, we assessed the expres-sion of DNTTIP1 and HDAC1 in HeLa cells arrested in mitosis or in early S-phase. No differences in expression were detected (Fig. 4a).

0

500

1,000

1,500

2,000

2,500

3,000

3,500

Contro

l IgG

DNTTIP1

(MiD

AC)

LSD1

(CoR

EST)

MTA3

(NuR

D)

SIN3A

(Sin3

)

RFUbNocodazoleNocodazole, TSA 10 µMVehicleVehicle, TSA 10 µM

*

Nocodazole

a

HDAC1

DNTTIP1

Probe-matrix boundHeLa extract

VehicleVehicle Aph. Noc. Aphidicolin

Antibody

Figure 4 Class I HDACs and DNTTIP1 form a mitotic deacetylase complex (MiDAC). (a) Cell cycle-dependent association of DNTTIP1 with the SAHA probe matrix was probed by western blot analysis. Lanes 1–3; identical expression levels of DNTTIP1 and HDAC1 in HeLa cells treated with aphidicolin (induces G1/S-phase arrest), nocodazole (induces arrest in mitosis) or vehicle. Lanes 4–9; increased amounts of DNTTIP1 is captured by the SAHA matrix from nocodazole-treated cells (lanes 6, 7) compared to aphidicolin (lanes 4, 5) or vehicle (lanes 8, 9). (b) Deacetylase activity assay of immunoaffinity-precipitated HDAC complexes demonstrates increased mitotic activity of a DNTTIP1-containing complex (MiDAC) but not of LSD1 (CoREST), MTA3 (NuRD) and SIN3A (Sin3)-containing complexes. Values are displayed as relative fluorescence units (RFU ± s.d.; N = 3; *P < 0.001 (Student’s t-test)). The fluorescence signal is reduced to background by 10 µM trichostatin A.

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnology  VOLUME 29 NUMBER 3 MARCH 2011 261

A rt i c l e s

Taken together with the fact that more DNTTIP1 was captured by the SAHA matrix from mitotic cells than from nonmitotic cells, the data indicated an increase in complex formation during mitosis, rather than an increase in the expression of constituent proteins. Next, we measured the deacetylase activity of the immunoprecipi-tated MiDAC complex (again isolated with the DNTTIP1 antibody), and compared it to the CoREST complex (LSD1 antibody), NuRD complex (MTA3 antibody) and Sin3 complex (SIN3A antibody). Whereas all samples possessed substantial deacetylase activity, only the MiDAC complex exhibited greater activity in division-arrested cells (Fig. 4b).

Cell-based profiling of HDAC inhibitorsTo investigate the correlation of proteomics target profiles with substrate selectivity, a subset of the reference inhibitors was sub-jected to cell-based tubulin and histone modification assays. K562 and HeLa cells were treated with vehicle or compounds, including nonselective inhibitors (SAHA, PCI-24781), class I selective inhibi-tors (tacedinaline, romidepsin, valproate) and the HDAC8 inhibitor

PCI-34051. Cell viability was monitored and drug effects were detected by antibodies for acetylated tubulin as the major substrate of HDAC6 (ref. 16) and histones H3 and H4 as the major class I HDAC substrates by immunofluorescence and western blot analysis (Fig. 5a and Supplementary Fig. 10).

The nonselective HDAC inhibitors increased steady-state acetyla-tion of tubulin and histones manifested by the staining of acetylated microtubules and punctuate nuclear staining of acetylated histones. The class I selective HDAC inhibitors stimulated histone acetylation but did not affect tubulin, as expected. Aliquots of vehicle-treated and drug-treated cells were also compared by differential map-ping of histone acetylation and methylation marks using quantita-tive high-resolution MS/MS33 (Fig. 5b and Supplementary Data Set 6). The results confirmed the range of activities observed in che-moproteomics profiling and indicate a pronounced abundance of hyperacetylated histone peptides after treatment with nonselective HDAC inhibitors, in particular TSA and romidepsin. In contrast, valproate exhibited a more selective effect, in particular less acetylation of H3K9 in peptides containing acetylated K14 (Fig. 5c).

Val

proa

teV

ehic

leS

AH

A

Overlay

PC

I−34

051

Tac

edin

alin

eP

CI−

2478

1R

omid

epsi

n

a Acetylated histone H3 Acetylated tubulin b c

Cmpd VC TSASAHA

PCI-247

81

Taced

inalin

e

Romide

psin

PCI-340

51

Bufex

amac

Valpro

ate

H2A 1 × ac

2 × ac

3 × ac

H2B 1 × ac

2 × ac

3 × ac

4 × ac

H3 1 × ac

2 × ac

H4

Histones

1 × ac

2 × ac

3 × ac

4 × ac

3.725Avg. log2 FC vs. unmod 0.000

0

2

4

0

2

4

0

2

4

Fol

d in

crea

se (

log2

sca

le)

Fol

d in

crea

se (

log2

sca

le)

Fol

d in

crea

se (

log2

sca

le)

KacSTGGKacAPR

KmeSTGGKacAPR

GKacGGKacGLGKacGGAKacR

Valpro

ate

Bufex

amac

PCI-340

51

Romide

psin

Taced

inalin

e

PC-247

81

SAHATSA

VC

Figure 5 Differential effects of HDAC inhibitors on histone and tubulin acetylation. Immunofluorescence analysis of histone H3 (K9ac/K14ac) and tubulin acetylation in HeLa cells treated for 4 h with vehicle, SAHA (10 µM), tacedinaline (50 µM), PCI-24781 (20 µM), PCI-34051 (100 µM), romidepsin (1 µM) or valproate (2mM). (a) Mapping of histone acetylation in K562 cells treated with HDAC inhibitors by LC-MS/MS. Cells were treated with TSA (10 µM), SAHA (5 µM), PCI-24781 (2 µM), tacedinaline (50 µM), romidepsin (1 µM), PCI-34051 (20 µM), bufexamac (100 µM) or valproate (2 mM) for 6 h. Histones were extracted from cells and acetylated peptides were quantified after isobaric tagging. (b) Heat map showing abundance of peptides with single or multiple acetylated lysines as dependent on inhibitor treatment. (c) Abundance of differently modified variants of the Histone H3.3 peptide 9-17 and the fully acetylated H4-peptide 5-19. Triplicate experiments were performed and error bars represent s.e.m. (Supplementary Data Set 6).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

262  VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

A rt i c l e s

Similarly, we observed fourfold less peracetylated H4(5-19) peptide in valproate-treated cells than in TSA-treated cells.

Bufexamac as a novel class IIb HDAC inhibitorTo discover novel selective HDAC inhibitors, we developed a high-throughput adaptation of the chemoproteomics protocol in which we replaced MS/MS detection with multiplexed fluorescent antibody detection on ‘dot blot’ arrays. The method was applied to the screening of a focused compound library in whole cell extracts of Jurkat and Ramos cells for inhibitors of HDAC1, 2, 3 and 6. Several hits were obtained with a few compounds display-ing a notable degree of selectivity (Fig. 6a). Two aminobenzamide fragments were identified as hit compounds exhibiting selectivity for HDAC3. Bufexamac, a nonsteroidal anti-inflammatory drug with an unknown mechanism of action34, preferentially affected HDAC6. Bufexamac was subjected to quantitative proteomics pro-filing as described above for the reference HDAC inhibitors set, which confirmed its selectivity for HDAC6 and HDAC10, the other class IIb isoform, in addition to several non-HDAC targets (Fig. 6b and Supplementary Data Set 2). The results are consistent with tubulin immunofluorescence and western blot data, which showed

a much larger amount of acetylated tubulin, the major HDAC6 substrate, but not of acetylated histones as substrates of class I HDACs (Figs. 5b and 6c). The cellular potency for tubulin deacetylation correlated with the potency for one of the drugs’ anti-inflammatory effects, the secretion of interferon (IFN)-α in peripheral blood mononuclear cells (Fig. 6d).

DISCUSSIONGene transcription and its epigenetic regulation are controlled by megadalton protein complexes35,36. Therefore, the action of drugs which modulate epigenetic mechanisms should be considered in the context of the multiprotein complexes they target. We developed an affinity capture method combined with multiplexed protein quantification by mass spectrometry to probe the interaction of drug molecules with drug targets in cells or tissue under conditions that preserve the integrity of protein complexes. To our knowledge, this is the first demonstration that small molecules exhibit different affinities toward (that is, they “recognize”) different protein complexes containing the same catalytic subunit. The strategy has the potential to be extended to other classes of pharmacological target, and enables the discovery of drug leads and their molecular targets as functional protein complexes. The biological

0.1 1.0 10 1000

50

100

a

c

b

dIC50 = 8.9 µM

IC50 = 2.9 µM

IFN

-α s

ecre

tion

[% in

hibi

tion]

Per

cent

inhi

bitio

n H

DA

C2

Percent inhibiton HDAC1

0 20 40 60 80 100

0

20

40

60

80

100

0.01 0.1 1.0 10 100

0

50

100

Viability

ac-H4K5

ac-tubulinac-H3K9/K14

0.01 0.1 1.0 10 100

0

50

100

Per

cent

inhi

bitio

n / v

iabi

lity

Per

cent

inhi

bitio

n / v

iabi

lity

Per

cent

inhi

bitio

n H

DA

C6

Percent inhibition HDAC1

0 20 40 60 80 100

0

20

40

60

80

100

Bufexamac

ONH

O

OH

Per

cent

inhi

bitio

n H

DA

C3

Percent inhibition HDAC1

0 20 40 60 80 100

0

20

40

60

80

100

Cpd AA-1Cpd AA-2

HN

O H2N

HN

N

O

H2N

AA

-2B

ufex

amac

Con

trol

[Bufexamac] (µM)

[Bufexamac] (µM)

[Bufexamac] (µM)

[Cpd AA-2] (µM)

Res

idua

l bin

ding

Acetylated histone H3K9 Acetylated tubulin Overlay

0

0.2

0.4

0.6

0.8

1.0

1.2

1 10 100 1,000

HDAC1 (NI)HDAC2 (NI)HDAC3(KD

app = 341 µM)

HDAC6(KD

app = 10.7 µM)

HDAC10 (KDapp

= 12.3 µM)HDAC8 (KD

app = 235 µM)

Figure 6 The nonsteroidal anti-inflammatory drug bufexamac is a novel class IIb HDAC inhibitor. (a) Screen of a focused compound library against HDACs 1, 2, 3 and 6 using a chemoproteomics binding assay with the SAHA matrix in whole cell extract from Jurkat and Ramos cells. The plots outline inhibition relative to HDAC1 for HDAC6, HDAC3 and HDAC2, as quantified by antibodies on dot-blot arrays. The compound concentration was 10 µM and chemical structures of selective hit compounds are shown. (b) HDAC selectivity profile of bufexamac in K562 cells, measured as outlined in Figure 1. (c) Treatment of HeLa cells with bufexamac elicits hyperacetylation of tubulin, whereas treatment with the o-aminoanilide AA-2 leads to hyperacetylation of histones. Cultured cells were treated with vehicle or drug for 4 h, and cells were analyzed by immunofluorescence microscopy and by western blot analysis using antibodies for acetylated tubulin (EC50 = 2.9 µM) and acetylated histones H3 (K9) and H4 (K5), respectively. (d) Treatment of peripheral blood mononuclear cells with bufexamac inhibits the secretion of IFN-α (EC50 = 8.9 ± 4.9 µM, three independent experiments).©

201

1 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature biotechnology  VOLUME 29 NUMBER 3 MARCH 2011 263

A rt i c l e s

activity of compounds is assessed without use of recombinant puri-fied proteins or protein overexpression. The major prerequisite is a probe matrix which binds to a sub-proteome which is characterized by a shared chemical ligand space, typically based on a substrate- or cofactor-binding site.

We developed a probe matrix that captures the class I and class IIb HDACs, and the majority of previously reported subunits of HDAC complexes, by binding to the substrate pocket. Class IIa HDACs were not identified, presumably because they exhibit low catalytic activity and low affinity for the hydroxamate probes14. Moreover, the probe matrix binds to many other enzymes which potentially represent tar-gets sharing a similar chemical ligand space, including other metallo-enzymes, as well as other proteins which may be associated with enzymes in protein complexes.

A basic application of a probe matrix is the differential expression profiling of a subproteome across a range of biological samples and conditions. In line with a pleiotropic function most proteins bind-ing to the hydroxamate matrix displayed minor differences across a panel of cell lines and tissues. However, two proteins lacking an obvi-ous small-molecule binding site were captured predominantly from mitotically arrested cells (DNTTIP1 and MIDEAS), in line with the formation of a specific mitotic HDAC complex. We confirmed and extended this finding by using inhibitor profiling, co-IP and enzyme assay data.

A powerful application of the probe matrix is the profiling of drugs and lead molecules interacting with proteins and protein complexes in cells. We demonstrated that robust quantitative data are obtained by high sensitivity liquid chromatography (LC)/LC-MS/MS to mea-sure protein binding to the matrix in whole cell extract as a function of the concentration of competing “free” inhibitor. Unbiased bidirec-tional hierarchical clustering of the proteomics target profiles of 16 inhibitors (Fig. 2b) guides (i) the classification of drugs in selectiv-ity clusters, which are predicted to exhibit similar pharmacological effects, (ii) the grouping of protein targets in chemical space, and (iii) the assignment of targets to protein complexes. In order to distinguish individual targets that share a structurally similar ligand binding space from proteins associated in a complex, the chemoproteomics clustering is correlated with co-IP mapping of endogenous complexes in the same cell type37. Our data set clusters 16 inhibitors in terms of their effects on 1,251 proteins that specifically interact with the probe matrix. The clustering of inhibitors is driven by the major chemo-types represented by hydroxamates and aminobenzamides, with sev-eral hydroxamate sub-clusters, and reveals an unexpected degree of selectivity for inhibitors previously perceived as nonselective25,38,39. However, much of the published data is inconsistent, raising issues with the enzyme assays employed. Notably, a recent carefully control-led extensive enzyme kinetic study of HDAC inhibitors also reported a higher degree of inhibitor selectivity14. Remarkably the compounds in the aminobenzamide cluster showed several distinctive features. Extending previous findings30, we observed slow binding to class I complexes, in particular HDAC1/2-dependent CoREST and NuRD complexes. Moreover, we found a characteristic selectivity profile with a preference for the HDAC3-NCoR complex and no or minor effects on the HDAC1/2-dependent Sin3 complex. It is tempting to speculate whether this selectivity profile may contribute to a more favorable toxicology profile or to reduced clinical efficacy40. Similarly, the clinically used drug valproate also affected the Sin3 complex to a lesser degree than other class I complexes. We did not find major differences when we assessed the effect of inhibitors with different selectivities on global histone acetylation. However, little is known about site specificity of the different complexes and their relative

cellular activities, which are likely cell-type and gene specific. To study these effects our methodology could be extended to include chromatin immunoprecipitation performed with the targets and anti-bodies validated in the co-IP studies.

A number of non-HDAC targets are potently affected by several hydroxamate HDAC inhibitors but do not appear to be compo-nents of HDAC complexes, given that their inhibition profiles do not match that of any HDAC (Fig. 2), and because they were not enriched in the co-IP set (Fig. 3a). These proteins may represent off-targets sharing a similar chemical ligand space. Examples are the basic leucine zipper/W2 domain protein BZW2 and the isochoris-matase domain protein ISOC2, and several other Zn2+-dependent metalloenzymes.

In the protein dimension, the clustering data delineate target pro-tein complexes, as proteins exhibiting matching inhibition profiles across the inhibitor panel are likely to be physically associated. This is evidenced by the excellent clustering of the four major HDAC-containing complexes (Fig. 2b). To our knowledge, this is the first time that small-molecule binding data are used to characterize target protein complexes. We thus extended our data by conducting an extensive co-IP analysis of endogenous HDAC complexes from the same cell extract. A few previously reported class I HDAC complex components that did not bind to the probe matrix were identified in the co-IP samples and may represent interactions that are sensi-tive to inhibitor binding, for example, the ING2 subunit of the Sin3 complex41. The co-IP results confirmed additional HDAC1/HDAC2 complexes delineated in the analysis of the chemoproteomics data. These complexes are built around ELM-SANT domain subunits that are phylogenetically related to corepressor components of NuRD and CoREST complexes31. Several of such complexes exist with each containing a single ELM-SANT scaffold, such as MIER1, MIER2, MIER3, RERE, TRERF1 or MIDEAS, a previously unannotated gene product with homology to the REST corepressor. One func-tion of HDAC complexes is likely the coordination of deacetylation with other epigenetic modifications. The CoREST complex couples HDACs to the demethylase LSD19, and MIER1 and RERE were shown to scaffold HDACs with the EHMT methyltransferases31, and our inhibition profiles confirmed these complexes as HDAC inhibitor targets. The composition of these HDAC complexes was deconvoluted further by the co-IP data, in particular the MiDAC complex formed by HDAC1/2, MIDEAS and DNTTIP1. DNTTIP1 is a DNA binding protein that has been described to modulate the activity of terminal deoxynucleotidyl transferase (TDT), a special-ized DNA polymerase that incorporates nontemplated nucleotides to the 3′ end of DNA templates to mediate the junctional diversity of immunoglobulin genes42. However, we did not consistently identify an association of TDT with the MiDAC complex, suggesting a TDT independent function of the complex in cell division. The inhibition profiles also implicated the REST corepressor CDYL32 as a compo-nent of MiDAC, in line with the increase in CDYL captured from mitotic cells by the SAHA matrix. CDYL co-immunopurified with HDAC2 but not with the MiDAC subunit DNTTIP1, and hence further analysis is required to clarify whether it is a component of MiDAC or of an alternative complex.

The fourth class I enzyme, HDAC8, is more difficult to assign to a complex, because it is only targeted by a few inhibitors. We did not identify a suitable antibody to characterize HDAC8 by co-IP. The HDAC8 inhibitor PCI-3405143 was the only compound in our panel that was specific for a single HDAC. The class IIb enzymes HDAC6 and HDAC10 were only inhibited by hydroxamate type compounds and both do not appear to form robust complexes as no

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

264  VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

A rt i c l e s

strong associations with other proteins in the inhibition profiles were detected. HDAC6 has recently been implicated in chromatin regula-tion44 but the HDAC6/HDAC10 inhibitor bufexamac did not affect the acetylation of histones, suggesting that class IIb deacetylases do not play a direct role of in histone modification.

Our chemoproteomics methodology can be adapted to high throughput screening by using an antibody-based readout to reduce sample requirements and process time. We conducted a screen of a focused compound library for selective inhibitors. The screen identi-fied the hydroxamate drug bufexamac, an NSAID with an unknown mechanism of action34 as a class IIb selective inhibitor. Its profile was unique among the set of inhibitors studied. The drug induced tubulin hyperacetylation in drug concentrations matching its anti- inflammatory effect. Therefore, inhibition of HDAC6 may contribute to the clinical efficacy of bufexamac.

In conclusion, we have shown that a chemoproteomics strategy based on small-molecule inhibitors can be applied to discover and classify molecular complexes around drug target proteins, which has not, to our knowledge, been previously shown. The approach con-firms and extends orthogonal protein-protein interaction mapping. We have demonstrated the utility of this strategy in drug discovery by measuring distinctive target profiles for clinical HDAC inhibitors in cell extracts, and employed it in screening for novel inhibitors. The data support the value of drug discovery strategies based on target proteins in their biological context.

METHODSMethods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/.

Accession numbers. PRIDE database (http://www.ebi.ac.uk/pride): mass spectrometry data set accession numbers 15345–15472.

Note: Supplementary information is available on the Nature Biotechnology website.

ACKNowlEDGMENTSThis work was supported by a grant from the German Bundesministerium für Bildung und Forschung (Spitzencluster BioRN, Verbundprojekt Inkubator/Teilprojekt INE-TP01) to Cellzome AG. We are grateful to N. Garcia-Altrieth, M. Jundt, M. Löttgers, J.-I. Huber, M. Klös-Hudak, J. Krause, B. Kröh, A. Podszuweit, T. Rudi and K. Weis for expert technical assistance, to C. Gemünd and V. Wolowski for the development of software and database tools, and to F. Weisbrodt for help with the figures. We would like to thank T. Edwards, O. Rausch and D. Simmons for suggestions and support.

AUTHoR CoNTRIBUTIoNSA.D., D.E., A.-M.M., and K.S. performed biochemical and cell biological experiments; V.R. synthesized and sourced compounds; D.P. performed the interferon assay; I.B. analyzed histone modifications; B.D., M.D. and M. Boesche prepared peptide samples and operated mass spectrometers; M. Bantscheff, M.M.S., T.M. and G.S. established and conducted mass spectrometry data handling processes; M.M.S., Y.A., C. Huthmacher and J.S. contributed data analysis and visualization; M. Bantscheff, C. Hopf, P.G. and G.D. analyzed data, planned and supervised experiments, and conceptualized the project; G.B., U.K., G.N. and N.G.R. contributed ideas and supported the work; and M. Bantscheff and G.D. wrote the paper.

CoMPETING FINANCIAl INTERESTSThe authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturebiotechnology/.

Published online at http://www.nature.com/naturebiotechnology/. reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/.

1. Kouzarides, T. Chromatin modifications and their function. Cell 128, 693–705 (2007).

2. Choudhary, C. et al. Lysine acetylation targets protein complexes and co-regulates major cellular functions. Science 325, 834–840 (2009).

3. Zhao, S. et al. Regulation of cellular metabolism by protein lysine acetylation. Science 327, 1000–1004 (2010).

4. Karberg, S. Switching on epigenetic therapy. Cell 139, 1029–1031 (2009).5. Taunton, J., Hassig, C.A. & Schreiber, S.L. A mammalian histone deacetylase related

to the yeast transcriptional regulator Rpd3p. Science 272, 408–411 (1996).6. Gregoretti, I.V., Lee, Y.M. & Goodson, H.V. Molecular evolution of the histone

deacetylase family: functional implications of phylogenetic analysis. J. Mol. Biol. 338, 17–31 (2004).

7. Yang, X.J. & Seto, E. The Rpd3/Hda1 family of lysine deacetylases: from bacteria and yeast to mice and men. Nat. Rev. Mol. Cell Biol. 9, 206–218 (2008).

8. Cunliffe, V.T. Eloquent silence: developmental functions of Class I histone deacetylases. Curr. Opin. Genet. Dev. 18, 404–410 (2008).

9. You, A., Tong, J.K., Grozinger, C.M. & Schreiber, S.L. CoREST is an integral component of the CoREST-human histone deacetylase complex. Proc. Natl. Acad. Sci. USA 98, 1454–1458 (2001).

10. Tong, J.K., Hassig, C.A., Schnitzler, G.R., Kingston, R.E. & Schreiber, S.L. Chromatin deacetylation by an ATP-dependent nucleosome remodelling complex. Nature 395, 917–921 (1998).

11. Zhang, Y., Iratni, R., Erdjument-Bromage, H., Tempst, P. & Reinberg, D. Histone deacetylases and SAP18, a novel polypeptide, are components of a human Sin3 complex. Cell 89, 357–364 (1997).

12. Karagianni, P. & Wong, J. HDAC3: taking the SMRT-N-CoRrect road to repression. Oncogene 26, 5439–5449 (2007).

13. Guenther, M.G., Barak, O. & Lazar, M.A. The SMRT and N-CoR corepressors are activating cofactors for histone deacetylase 3. Mol. Cell. Biol. 21, 6091–6101 (2001).

14. Bradner, J.E. et al. Chemical phylogenetics of histone deacetylases. Nat. Chem. Biol. 6, 238–243 (2010).

15. Lahm, A. et al. Unraveling the hidden catalytic activity of vertebrate class IIa histone deacetylases. Proc. Natl. Acad. Sci. USA 104, 17335–17340 (2007).

16. Boyault, C., Sadoul, K., Pabion, M. & Khochbin, S. HDAC6, at the crossroads between cytoskeleton and cell signaling by acetylation and ubiquitination. Oncogene 26, 5468–5476 (2007).

17. Marks, P.A. & Breslow, R. Dimethyl sulfoxide to vorinostat: development of this histone deacetylase inhibitor as an anticancer drug. Nat. Biotechnol. 25, 84–90 (2007).

18. Bolden, J.E., Peart, M.J. & Johnstone, R.W. Anticancer activities of histone deacetylase inhibitors. Nat. Rev. Drug Discov. 5, 769–784 (2006).

19. Zhang, Y. et al. Analysis of the NuRD subunits reveals a histone deacetylase core complex and a connection with DNA methylation. Genes Dev. 13, 1924–1935 (1999).

20. Salisbury, C.M. & Cravatt, B.F. Activity-based probes for proteomic profiling of histone deacetylase complexes. Proc. Natl. Acad. Sci. USA 104, 1171–1176 (2007).

21. Bantscheff, M. et al. Quantitative chemical proteomics reveals mechanisms of action of clinical ABL kinase inhibitors. Nat. Biotechnol. 25, 1035–1044 (2007).

22. Ong, S.E. et al. Identifying the proteins to which small-molecule probes and drugs bind in cells. Proc. Natl. Acad. Sci. USA 106, 4617–4622 (2009).

23. Sharma, K. et al. Proteomics strategy for quantitative protein interaction profiling in cell extracts. Nat. Methods 6, 741–744 (2009).

24. Bantscheff, M., Scholten, A. & Heck, A.J. Revealing promiscuous drug-target interactions by chemical proteomics. Drug Discov. Today 14, 1021–1029 (2009).

25. Khan, N. et al. Determination of the class and isoform selectivity of small-molecule histone deacetylase inhibitors. Biochem. J. 409, 581–589 (2008).

26. Bantscheff, M. et al. Robust and sensitive iTRAQ quantification on an LTQ Orbitrap mass spectrometer. Mol. Cell. Proteomics 7, 1702–1713 (2008).

27. Jones, P. et al. Probing the elusive catalytic activity of vertebrate class IIa histone deacetylases. Bioorg. Med. Chem. Lett. 18, 1814–1819 (2008).

28. Kruhlak, M.J. et al. Regulation of global acetylation in mitosis through loss of histone acetyltransferases and deacetylases from chromatin. J. Biol. Chem. 276, 38307–38319 (2001).

29. Savitski, M.M. et al. Targeted data acquisition for improved reproducibility and robustness of proteomic mass spectrometry assays. J. Am. Soc. Mass Spectrom. 21, 1668–1679 (2010).

30. Chou, C.J., Herman, D. & Gottesfeld, J.M. Pimelic diphenylamide 106 is a slow, tight-binding inhibitor of class I histone deacetylases. J. Biol. Chem. 283, 35402–35409 (2008).

31. Wang, L., Charroux, B., Kerridge, S. & Tsai, C.C. Atrophin recruits HDAC1/2 and G9a to modify histone H3K9 and to determine cell fates. EMBO Rep. 9, 555–562 (2008).

32. Mulligan, P. et al. CDYL bridges REST and histone methyltransferases for gene repression and suppression of cellular transformation. Mol. Cell 32, 718–726 (2008).

33. Savitski, M.M., Mathieson, T., Becher, I. & Bantscheff, M. H-score, a mass accuracy driven rescoring approach for improved Peptide identification in modification rich samples. J. Proteome Res. 9, 5511–5516 (2010).

34. Trommer, H. et al. Examinations of the antioxidative properties of the topically administered drug bufexamac reveal new insights into its mechanism of action. J. Pharm. Pharmacol. 55, 1379–1388 (2003).

35. Alberts, B. The cell as a collection of protein machines: preparing the next generation of molecular biologists. Cell 92, 291–294 (1998).

36. Gavin, A.C. et al. Proteome survey reveals modularity of the yeast cell machinery. Nature 440, 631–636 (2006).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnology  VOLUME 29 NUMBER 3 MARCH 2011 265

A rt i c l e s

37. Malovannaya, A. et al. Streamlined analysis schema for high-throughput identification of endogenous protein complexes. Proc. Natl. Acad. Sci. USA 107, 2431–2436 (2010).

38. Beckers, T. et al. Distinct pharmacological properties of second generation HDAC inhibitors with the benzamide or hydroxamate head group. Int. J. Cancer 121, 1138–1148 (2007).

39. Blackwell, L., Norris, J., Suto, C.M. & Janzen, W.P. The use of diversity profiling to characterize chemical modulators of the histone deacetylases. Life Sci. 82, 1050–1058 (2008).

40. Farias, E.F. et al. Interference with Sin3 function induces epigenetic reprogramming and differentiation in breast cancer cells. Proc. Natl. Acad. Sci. USA 107, 11811–11816 (2010).

41. Smith, K.T., Martin-Brown, S.A., Florens, L., Washburn, M.P. & Workman, J.L. Deacetylase inhibitors dissociate the histone-targeting ING2 subunit from the Sin3 complex. Chem. Biol. 17, 65–74 (2010).

42. Kubota, T., Maezawa, S., Koiwai, K., Hayano, T. & Koiwai, O. Identification of functional domains in TdIF1 and its inhibitory mechanism for TdT activity. Genes Cells 12, 941–959 (2007).

43. Balasubramanian, S. et al. A novel histone deacetylase 8 (HDAC8)-specific inhibitor PCI-34051 induces apoptosis in T-cell lymphomas. Leukemia 22, 1026–1034 (2008).

44. Wang, Z. et al. Genome-wide mapping of HATs and HDACs reveals distinct functions in active and inactive genes. Cell 138, 1019–1031 (2009).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnology doi:10.1038/nbt.1759

ONLINE METHODSReagents. All reagents were purchased from Sigma unless otherwise noted below. Antibodies were purchased from the following suppliers: sc-7872 (HDAC1), sc-81599 (HDAC2), sc-17795 (HDAC3), sc-11405 (HDAC8), sc-81325 (MTA3), sc-100908 (TBL1XR1), sc-81082 and sc-166296 (DNTTIP1) and sc47778 (β-actin) from Santa Cruz; 05-814 (HDAC2), 05-813 (HDAC3), 07-505 (HDAC8) from Millipore; ab46985 (HDAC1), ab3479 (Sin3A), ab70039 (DNTTIP1); NB100-40825 (EHMT2), NB100-81655 (TRERF1) and NB100-81654 (TRERF1) from Novus Biologicals; ab5188 (CDYL) and ab61236 (H4-AcK5) from Abcam; H-3034 (HDAC-3) and T-6793 (Ac-tubulin) from Sigma; H00009425-M02 (CDYL) and H00010013-M01 (HDAC6) from Abnova; no. 2184 (LSD1) from Cell Signaling Technologies; and 382158 (H3-AcK9/K14) from Calbiochem. Secondary antibodies labeled with IRDye 680 and IRDye 800 were from LICOR, and antibodies labeled with Alexa 480 and Alexa 594 were from Invitrogen. The HDAC activity assay was purchased from ActiveMotif. Reference compounds were purchased from the following suppliers: vorinostat (SAHA), belinostat (PXD-101), dacinostat (LAQ-824), panobinostat (LBH-589), PCI-24781, and entinostat (MS-275) from Selleck; trichostatin A, PCI-34051, bufexamac and apicidin from Sigma; scriptaid and tacedinaline (CI-994) from Tocris; MC-1293 and BML-210 from Enzo Life Sciences; MGCD-0103 from Chemietek; romidepsin (FK-228) from ACC Corp; and valproic acid from Calbiochem. All other compounds were syn-thesized as described in Supplementary Synthetic Procedures.

Cell culture. Jurkat E6.1, HL60, Ramos and HeLa cells were purchased from American Type Culture Collection; K562 cells were purchased from DSMZ. Jurkat E6.1 cells were cultured in RPMI1640 supplemented with 4.5 g/l glucose, 10 mM HEPES, 1 mM sodium pyruvate and 10% FCS. Ramos cells were cultured in RPMI1640 containing 10% FCS. K562 cells were cul-tured in RPMI medium containing 10% FCS. Cells were expanded to maximal 1 × 106 cells/ml. HeLa cells were cultured in minimum essential media (MEM) supplemented with 1 mM pyruvate, 0.1 mM nonessential amino acids and 10% FCS. For indirect immunofluorescence assays, the FCS content was reduced to 2%. For cell cycle arrest of HeLa cells in G1/S phase or mitosis, cells were treated for 16 h with 15 µg/ml aphidicolin (Sigma) or with 0.3 µM nocodazole (Sigma). Control HeLa cells were treated with DMSO for 16 h.

Preparation of cell lysates. Frozen cell pellets were homogenized in lysis buffer (50 mM Tris-HCl, 0.8% Igepal-CA630, 5% glycerol, 150 mM NaCl, 1.5 mM MgCl2, 25 mM NaF, 1 mM sodium vanadate, 1 mM DTT, pH 7.5). One complete EDTA-free protease inhibitor tablet (Roche) per 25 ml was added. The sample was dispersed using a Dounce homogenizer, kept rotating for 30 min at 4 °C and spun for 10 min at 20,000g at 4 °C. The supernatant was spun again for 1 h at 145,000g. The protein concentration was determined by Bradford assay (BioRad), and aliquots were snap frozen in liquid nitrogen and stored at −80 °C.

Proteomics-based inhibitor profiling. Each inhibitor profiling experiment is denoted by an experiment identifier (internal diameter) number. A list of all profiling experiments is provided in Supplementary Table 5 including all experimental parameters like preincubation times, inhibitor concentrations, isobaric tagging schemata and MS method used.

Affinity profiling assays were carried out as described previously21 with minor modifications. Derivatized sepharose beads (35 µl beads per sample) were equili-brated in lysis buffer and incubated with 1 ml (5 mg protein) cell lysate, which had been preincubated with test compound or vehicle for 45 min, on an end-over-end shaker for 1 h. Incubation was done at 4 °C for all compounds. In addition, experiments were also performed at 22 °C for the aminobenzamide compounds (tacedinaline, entinostat, BML-120 and mocetinostat). In some experiments different preincubation times (0 to 240 min) or temperatures (4 °C or 22 °C) were tested, to assess the influence of binding kinetics on the selectivity profiles. Beads were transferred to disposable columns (MoBiTec), washed with lysis buffer containing 0.2% NP-40 and eluted with 50 µl 2× SDS sample buffer. Proteins were alkylated with 200 mg/ml iodoacetamide for 30 min, separated on 4–12% NuPAGE (Invitrogen), and stained with colloidal Coomassie.

Compound screening. The compounds for screening were selected either for their potential to be zinc chelators or based on a similarity search of the

compound collections from Asinex (http://www.asinex.com/chemsearch.html) and Enamine (http://www.enamine.net/), using a training set of 140 known HDAC inhibitors. Competition binding assays using the SAHA matrix were done essentially as described above but adapted to a 96-well format. We used 1mg of cell lysate and 5 µl of beads per well. Compounds from the screening library including reference compounds as standards were added at 20 µM and 100 µM final concentration from 50× DMSO stocks. Each plate con-tained eight positive (TSA, 50 µM) and eight negative controls (2% DMSO). Beads were eluted in SDS sample buffer (100 mM Tris pH 7.4, 4% SDS, 20% glycerol, 0.01% bromophenol blue, 50 mM DTT) and spotted in duplicate on nitrocellulose membranes (600 nl/spot) using an automated liquid dispenser (Fluid). After drying, the membranes were rehydrated in 20% ethanol, and processed for detection with specific antibodies as indicated. Spot intensities were quantified using a LiCOR Odyssey scanner and percentage inhibition was calculated using positive and negative controls as 100% and 0% inhibi-tion, respectively.

Quantitative co-IP. Antibodies were tested for suitability in co-IP assays by immunoprecipitation-western blot analysis procedures. For western blot analysis we used a LI-COR Odyssey System. Suitable antibodies (40–100 µg) were coupled to 100 µl AminoLink resin (Thermo Fisher Scientific). Cell lysate samples (10 mg) were incubated with prewashed immuno resin on a shaker for 2 h at 4 °C. Beads were washed in lysis buffer containing 0.4% Igepal-CA630 and lysis buffer without detergent. Bound proteins were eluted in 100 µl 2× SDS sample buffer. Protein samples were reduced, alkylated and separated by SDS-PAGE. To provide a specificity control for quantita-tive LC-MS analysis, IgG from the same species was used in an analogous ‘mock IP’ carried out in parallel from an aliquot of the same lysate sample. Typically, four IP reactions, which were subsequently combined in a single iTRAQ sample for MS/MS analysis, were done in parallel, two with differ-ent antibodies directed against the same (or different) antigen(s) and two ‘mock IP’ samples.

Enzymatic deacetylation assays. The enzyme activity of purified recom-binant HDAC1, HDAC2, HDAC3-NCoR and HDAC6 (1 µg of protein per well) was measured using the Active Motif HDAC Assay Kit in 96-well for-mat following the manufacturer’s instructions. Fluorescence measurements (340 nm excitation/460 nm emission) were recorded with an Analyst HT plate reader (Molecular Devices) in triplicates. Each time series was performed in duplicate. For the determination of enzymatic activity in endogenous HDAC complexes, cell extract samples (375 µl each at 4 mg/ml protein concentra-tion) were prepared from either nocodazole- or vehicle-treated HeLa cultures and were incubated with 15 µg of each antibody for 2 h at 4 °C. Subsequently, 60 µl Protein G beads equilibrated in lysis buffer containing 0.4% NP40 were added and, following incubation for 1 h at 4 °C, beads were washed twice with lysis buffer followed by 10 volumes of Assay buffer (HDAC Assay Kit). We added 10 µl of beads per well and deacetylation activity was determined as described above after 150 min in the absence or presence of 10 µM trichostatin A. Each point was measured in triplicate. Statistical significance was assessed by unpaired Student’s t-test.

Sample preparation for MS. Gels were cut into slices across the entire sep-aration range and subjected to in-gel digestion21. For acquisition of dose-response inhibitor data in one single multiplexed run, TMT (Thermo-Fisher Scientific) tags were used because they allow the acquisition of six-point data. For immunoaffinity purifications, a maximum of four samples was compared, and iTRAQ reagents (Applied Biosystems) were used for reasons of economy and coverage45. Peptide extracts were labeled with iTRAQ or TMT in 40 mM triethylammoniumbicarbonate, pH 8.53. After quenching of the reaction with glycine, labeled extracts were combined. For compound profiling experiments extracts from vehicle-treated samples were labeled with TMT reagent 131, and combined with extracts from compound-treated samples labeled with TMT reagents 126–130, fractionated using reversed-phase chromatography at pH 12, dried and acidified before LC-MS/MS analysis.

LC-MS/MS analysis. Samples were dried in vacuo and resuspended in 0.1% formic acid in water and aliquots of the sample were injected into a

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnologydoi:10.1038/nbt.1759

nano-LC system (Eksigent 1D+) coupled to LTQ-Orbitrap mass spectrometers (Thermo-Finnigan). Peptides were separated on custom 50 cm × 75 µM (inter-nal diameter) reversed-phase columns (Reprosil) at 40 °C. Gradient elution was performed from 2% acetonitrile to 40% acetonitrile in 0.1% formic acid over 2–3 h. LTQ-Orbitrap XL and Orbitrap Velos instruments were operated with XCalibur 2.0/2.1 software. Intact peptides were detected in the Orbitrap at 30.000 resolution. Internal calibration was performed using the ion signal from (Si(CH3)2O)6H+ at m/z 445.120025 (ref. 46). Data-dependent tandem mass spectra were generated for up to six peptide precursors using a com-bined CID/HCD approach47 or using HCD at a resolution of 7,500 for histone modification data. For CID, up to 5,000 ions (Orbitrap XL) or up to 3,000 ions (Orbitrap Velos) were accumulated in the ion trap within a maximum ion accumulation time of 200 msec. For HCD, target ion settings were 50,000 (Orbitrap XL) and 25,000 (Orbitrap Velos), respectively.

Peptide and protein identification. Mascot 2.0 (Matrix Science) was used for protein identification using 10 p.p.m. mass tolerance for peptide pre-cursors and 0.8 Da (CID) or 20 mDa (HCD) tolerance for fragment ions. Carbamidomethylation of cysteine residues and iTRAQ/TMT modification of lysine residues were set as fixed modifications and S,T,Y phosphorylation, methionine oxidation, N-terminal acetylation of proteins and iTRAQ/TMT modification of peptide N termini were set as variable modifications. The search database consisted of a customized version of the International Protein Index database combined with a decoy version of this database created using a script supplied by Matrix Science. Unless stated otherwise, we accepted protein identifications as follows: (i) for single spectrum to sequence assignments, we required this assignment to be the best match and a minimum Mascot score of 31 and a 10× difference of this assignment over the next best assignment. Based on these criteria, the decoy search results indicated <1% false-discovery rate (FDR); (ii) for multiple spectrum to sequence assignments and using the same parameters, the decoy search results indicate <0.1% FDR. For protein quantification a minimum of two sequence assignments matching to unique peptides was required. FDR for quantified proteins was <<0.1%. For analysis of histone peptides additional variable modifications considered were acetylation of lysine, mono- di- and trimethylation of lysine, mono- and dimethylation of arginine. Localization of post-translational modifications on histone peptides was validated by remapping the de-isotoped and de-convoluted tandem MS spectra to b and y ions expected from the peptide hit within 20 p.p.m. mass accuracy. Only spectrum to sequence matches for which fragment ions support localization on only one amino acid and that yielded an H score > 6 indicating 99% confidence33 were considered for further analysis29. For identification of peptides with acetylated lysine the presence of the 126.0913 immonium ion was required.

Peptide and protein quantification. Centroided iTRAQ/TMT reporter ion signals were computed by the XCalibur software operating and extracted from MS data files using customized scripts. Only peptides unique for iden-tified proteins were used for relative protein quantification and are referred to in Supplementary Data Sets 1–6 as unique peptide assignments (UPA). Further, quantification spectra matching to UPAs were filtered according to the following criteria: mascot ion score > 15, signal to background ratio of the precursor ion > 4, signal to interference > 0.5 (ref. 29). Reporter ion intensities were multiplied with the ion accumulation time yielding an area value proportional to the number of reporter ions present in the mass analyzer. For compound competition binding experiments, fold-changes are reported based on reporter ion areas in comparison to vehicle control and were calculated using sum-based bootstrap algorithm. Fold-changes were corrected for isotope purity as described and adjusted for interference caused by co-eluting nearly isobaric peaks as estimated by the signal-to- interference measure29.

The heat maps in Supplementary Figure 2a–d are based on the accumu-lated reporter ion responses for each protein divided by its molecular weight (10 percentile bins). This enables accurate relative quantification between differ-ent conditions for the same protein and similarly to spectrum count based meth-ods gives an estimation of the relative abundance across different proteins48. Fractional abundance was calculated by the reporter ion response in condition i divided by the summed reporter ion response across all conditions:

FA i A i

A jjn( ) ( )

( )=

=∑ 1

Dose-response curves were fitted using R (http://www.r-project.org/) and the drc package (http://www.bioassay.dk), as described previously21. IC50 values were confirmed in replicate experiments using targeted data acquisi-tion for a subset of proteins29. To compare selectivities of compounds dis-playing different absolute potencies relative potencies were calculated as (pIC50 – min(pIC50))/(max(pIC50) – min(pIC50)) for each experiment. Apparent dissociation constants (Kd

app) were derived from IC50 values as described23 and relative affinities were determined as described for relative potencies. For the assessment of data robustness (Supplementary Fig. 4), seven replicate samples of trichostatin A treated samples were analyzed as pre-viously described. IC50 values for 21 proteins were calculated for each experi-ment and normalized using the average pIC50 determined in each experiment. Then average and s.d. of the pIC50 for each protein across the seven samples were calculated. A pair-wise, two sided t-test was performed between pIC50 values of each protein with those of every other protein. In addition P-values, average pIC50, and differences between pIC50 were calculated after combining the proteins into complexes.

For immunoprecipitations, the enrichment E was calculated as (A(IP) – A(mock IP))/(A(IP) + A(mock IP)) and scales between −1 and 1. ‘A’ represents the summed-up reporter ion response for the protein of interest. When immunoprecipitation experiments against two different bait proteins (IP1, IP2) were analyzed in a single iTRAQ experiment, relative enrichment (Fig. 3a) was calculated as RE(IP1) = (A(IP1) – A(mock IP))/(A(IP1) + A(IP2) + A(mock IP)). RE of 0.5 means that the protein was precipitated in both IPs equally well if no signal was detected in the mock IP.

C score–based determination of complex components using compound profiling and immunoprecipitation data sets. For each protein identified in an immunoprecipitation experiment against one of the bait proteins (7,000 unique bait/protein pairs) enrichment was normalized to values between 0 and 1. Similarly average pIC50s retrieved from compound profiling experi-ments were normalized to values between 0 and 1 (5,683 unique proteins). The average normalized pIC50 (anpIC50) value was calculated across all com-pounds included for each protein. The anpIC50 of each protein is linked to the normalized enrichment (nPD) values determined in each immunoprecipita-tion experiment. The C-score for the 6,263 unique bait/protein combinations observed in compound profiling and immunoprecipitation experiments was calculated as the sum of squares of the anpIC50 and the nPD values. The result-ing value we dubbed C-score and it scales from 0 to 2. Similarly, C-score values for individual compound/bait pairs were calculated.

We tested the discrimination power of the C-score by scrambling all anpIC50 values and reassigning them to the 5,683 unique proteins. After that the linking of competition data and the pull-down data (again a total of 6,263 matches) as well as the C-score calculation were performed exactly as above.

In a next step we used this randomized C-score data set to determine the significance threshold for separating random hits from experimental data-bases on a cumulative FDR. For each C-score value X, the ratio of number of bait/protein pairs from the random matching and from the experimental matching that have a C-score ≥ X were calculated. This enabled us to determine the subset of high confidence interactions (FDR < 0.05: C-score > 1.14; FDR < 0.15: C-score >1.0 and max(c-score) > 1.2) (Supplementary Fig. 9).

Heat map generation. Heat maps and t-tests were performed using the R-package and Tableau.

For unbiased hierarchical clustering of compound profiling data, all quan-tified proteins identified with at least 9 independent experiments were con-sidered. Protein and compound clustering was based on relative affinities averaged over replicate experiments using the Euclidean distance measure and the complete linkage method provided in R. Only those proteins were considered for which binding to the SAHA matrix was inhibited by at least one compound (that is, an IC50 value was determined) and that were identi-fied in experiments related to more than half of the tested compounds. The significance of observed clusters was tested with the R package pvclust, which calculates two P-values for each cluster on the basis of a bootstrap resampling

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.

nature biotechnology doi:10.1038/nbt.1759

technique: approximately unbiased P-value (AU) and bootstrap probability value (BP)49. Clusters were reported as significant for AU > 95 indicating 95% probability of not being a random cluster.

Cell-based assays. For cell-based protein acetylation assays, HeLa or K562 cells (96-well format, 5 × 104 cells per well) were treated with compounds for 6 h. Cells were washed with cold PBS and lysed directly in SDS-sample buffer, followed by denaturation at 95 °C. Lysates (10 µl) were resolved on SDS-gels, transferred to PVDF membranes and analyzed for tubulin and histone acetylation by immune-detection using IRDye-labeled secondary antibodies and an Odyssey scanner (LiCOR). Data analysis of the quanti-fied bands was performed using Excel and GraphPad Prism. For the IFN-α secretion assay, human peripheral blood mononuclear cells were isolated with Histopaque 1077 (Sigma) from fresh human donor blood and were plated at a concentration of 0.5 × 106 cells/ml. Cells were incubated with test compounds for 45 min before stimulation with plasmacytoid dendritic cell–specific TLR9 agonist ODN2216 (Invivogen). IFN-α released into cell supernatants was measured in triplicate after 16 h by Flex-Set IFNalpha (BD Biosciences) by flow cytometry (FACSCalibur, BD Biosciences). For all cell-based assays viability was assessed in parallel (MTT kit, Roche). For histone modification experiments 1E6 K562 cells were treated with HDAC inhibitors for 6 h. After cell harvest, samples were subjected to histone enrich-ment as previously described50 and separated using SDS gel electrophoresis. For indirect immunofluorescence analysis, HeLa cells were plated to subcon-fluency on polylysine-coated glass chamber slides and treated after recovery

with the indicated compounds for 4 h. Samples were fixed in cold methanol, permeabilized with Triton-X100, blocked in 1% BSA and treated for immu-nodetection of acetylated tubulin and acetylated histone H3. Cells were coun-terstained for nucleic acids using 4′,6-diamidino-2-phenylindole (DAPI). Multichannel fluorescence microscopy was performed on an Olympus IX-70 microscope. Images were acquired using a monochrome CCD camera (CoolSNAP HQ Digital) and analyzed with MetaMorph (Universal Imaging Corporation). The instrument was adjusted to ensure proper comparison of levels of acetylated tubulin and acetylated histone H3 before and after inhibitor treatment.

45. Thingholm, T.E., Palmisano, G., Kjeldsen, F. & Larsen, M.R. Undesirable charge-enhancement of isobaric tagged phosphopeptides leads to reduced identification efficiency. J. Proteome Res. 9, 4045–4052 (2010).

46. Olsen, J.V. et al. Parts per million mass accuracy on an Orbitrap mass spectrometer via lock mass injection into a C-trap. Mol. Cell. Proteomics 4, 2010–2021 (2005).

47. Kocher, T. et al. High precision quantitative proteomics using iTRAQ on an LTQ Orbitrap: a new mass spectrometric method combining the benefits of all. J. Proteome Res. 8, 4743–4752 (2009).

48. Sanders, S.L., Jennings, J., Canutescu, A., Link, A.J. & Weil, P.A. Proteomics of the eukaryotic transcription machinery: identification of proteins associated with components of yeast TFIID by multidimensional mass spectrometry. Mol. Cell. Biol. 22, 4723–4738 (2002).

49. Suzuki, R. & Shimodaira, H. Pvclust: an R package for assessing the uncertainty in hierarchical clustering. Bioinformatics 22, 1540–1542 (2006).

50. Shechter, D., Dormann, H.L., Allis, C.D. & Hake, S.B. Extraction, purification and analysis of histones. Nat. Protoc. 2, 1445–1457 (2007).

© 2

011

Nat

ure

Am

eric

a, In

c. A

ll ri

gh

ts r

eser

ved

.