Chemoproteomics profiling of HDAC inhibitors ...

Articles

Chemoproteomics profiling of HDAC inhibitors reveals selective targeting of HDAC complexes

© 2011 Nature America, Inc. All rights reserved.

Marcus 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. Protein lysine acetylation is a key mechanism in the epigenetic control 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 complexes 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 developmental 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 ­s caffold ­f unctions14,15. 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 lymphoma, 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 selectivity. 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 activity 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.

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. ([email protected]) or G.D. ([email protected]).

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

nature biotechnology  VOLUME 29  NUMBER 3  MARCH 2011

255

Articles consist of a moiety that binds to a ligand pocket conserved within the target class under investigation, and a functional group for immobilization 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 derivatizing sepharose with analogs of the hydroxamates SAHA and givinostat (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 treatment was quantified by isobaric tagging of tryptic peptides and tandem 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

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 binding 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.

a

1

O

Probe matrix

O

H N

2

Vehicle control

3

NH

O N H

O

OH

O

4 TMT 131

OH

N H

5

6

Protein does not bind to drug

b

100

50

IP of protein A

6

m/z

12

12

7 12 8 12 9 13 0 13 1

0

TMT 130

Protein A binds to target 100

Protein B binds to target 100

Protein A Protein B

50

TMT 129

50

7 12 8 12 9 13 0 13 1

m/z

log [inhibitor]

12

12

log [inhibitor]

m/z

0

6

12

6 12 7 12 8 12 9 13 0 13 1

0

Target binds to drug Target

100

TMT 128 50

log [inhibitor]

6 12 7 12 8 12 9 13 0 13 1

0 12

Inhibitor concentration

TMT 127

m/z

Protein C binds to target Protein D binds to target

100

100

Protein C Protein D

50 50 log [inhibitor]

6 12 7 12 8 12 9 13 0 13 1

9 13 0 13 1

12

8

7 12

12

log [inhibitor]

m/z

0

6

12

TMT 126

0

12

© 2011 Nature America, Inc. All rights reserved.

RESULTS Synthesis of a target class–specific HDAC probe matrix Target class–directed chemical probes provide tools for the identification of drug targets directly in cells and tissues. Typically, probes

m/z

IP of protein D

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.

256

VOLUME 29  NUMBER 3  MARCH 2011  nature biotechnology

Articles a

NH2

Aminobenzamide

Tacedinaline (CI-994) O O

N H N H

NH2

NH

O

NH S

N H

O S

O O

O NH

10

0.01

0.1

1

10

0.01

0.1

1

10

100

99

0 1

Residual binding

© 2011 Nature America, Inc. All rights reserved.

O

1

1.0

10

100 1,000

1

10

100 1,000

1

10

100 1,000

1.0 95

1 0 1

Romidepsin

0.1

10

100 1,000

1

10

100 1,000

1.0

0

HDAC1 HDAC2 HDAC3 HDAC6 HDAC8 HDAC10

0.001 0.01

1

GATAD2A LSD1 MTA3 NCOR1 RCOR1 SIN3A TBL1XR1

0.1

1

[Compound] (µM)

0.001 0.01

0.1

1

10

100 1,000

BZW2 DNTTIP1 ISOC2 MIDEAS PDXK

0.001 0.01

[Compound] (µM)

0.1

1

[Compound] (µM)

0

PC I− 3 Va MC 40 lp −1 51 PCroic 293 R I−2 aci o d Ta mi 478 ce de 1 d ps En ina in tin line M BM ost oc L− at et 21 in Sc os 0 rip tat ta Be SA id lin HA Pa A os no pic tat b i D in di Tr ac os n ic in tat ho os st ta at t in A

N H

O

0 0.01

SAP30 SIN3A SAP30L SIN3B WIZ BRMS1 HDAC6 LOC153364 ISOC2 TBL1XR1 TBL1X NCOR2 NCOR1 GPS2 HDAC3 RCOR1 HMG20A LSD1 GSE1 RCOR3 HMG20B RCOR2 HDAC1 HDAC2 MIER3 MTA3 MTA1 MBD3 MIER2 PHF21A GATAD2B GATAD2A RBBP4 MTA2 ZMYM2 DNTTIP1 CDYL MIDEAS RREB1 MIER1 ZMYM3 RERE EHMT2 MBD2 CHD4 CDK2AP1

CoREST

O

H N

98

NCOR1

BML210

b

Other proteins

Relative potency

Hydroxamic acid

Residual binding

OH

Residual binding

N H

O

Complex subunits

1.0

SIN3

O

H N

Residual binding

HDACs

SAHA

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 pKdapp 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).

across a set of drugs. We confirmed associations of proteins in complexes using immunoaffinity purifications (Fig. 1). Quantitative mapping of protein binding to the probe matrix We 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 activity 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 proteins identified, 267 proteins exhibited substantially reduced matrix nature biotechnology  VOLUME 29  NUMBER 3  MARCH 2011

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, implicating 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 captured 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 differentially 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 division 28, we conducted differential mapping of the subproteome in HeLa cells arrested in mitosis 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, 257

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

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 statistically significant differences between distinct com­plexes ­containing

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, including 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

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

© 2011 Nature America, Inc. All rights reserved.

Romidepsin Tacedinaline BML-210 SAHA preincub. Trichostatin A Valproic acid Belinostat ­ reincub. at 4 °C preincub. at 22 °C ­preincub. at 4 °C at 4 °C max conc. ­preincub. at 22 °C ­preincub. at 4 °C ­preincub. at 4 °C p 10 µM max conc. 30 µM max conc. 300 µM max conc. 100 nM max conc. 600 µM max conc. 1 µM max conc. 20 mM Classification

Protein

Kdappa

s.d.b

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Class I HDAC

HDAC1 HDAC2 HDAC3 HDAC8

0.29 0.35 0.26 >30c

0.02 0.04 0.08 –

14.9 10.7 2.8 > 300

1.0 0.2 0.4 –

0.008 0.010 0.017 > 0.1

0.001 0.001 0.001 –

0.28 0.35 0.29 > 10

0.05 0.08 0.06 –

13.0 11.9 6.0 240

1.8 1.4 0.3 137

0.010 0.012 0.041 >1

0.001 0.002 0.007 –

745 975 9,350 12,675

147 249 3949 734

Class II HDAC

HDAC6 HDAC10

0.53 >30

0.14 –

> 300 > 300

– –

> 0.1 > 0.1

– –

0.13 5.26

0.04 2.34

> 600 > 600

– –

0.109 >1

0.025 >20,000 – >20,000

CoREST complex

GSE1 LSD1 HMG20A HMG20B PHF21A RCOR1 RCOR2 RCOR3 ZMYM2 ZMYM3

0.29 0.29 0.31 0.33 0.25 0.20 0.21 0.26 0.36 0.39

0.09 0.07 0.09 0.10 0.04 0.03 0.06 0.08 0.06 0.24

9.2 9.2 9.8 12.6 31.8 8.5 18.2 10.5 43.3 134

0.9 0.5 0.4 0.5 1.0 0.5 5.5 0.3 32.2 90.9

0.006 0.006 0.005 0.006 0.012 0.005 0.008 0.005 0.013 0.035

0.001 0.001 0.001 0.001 – 0.001 0.000 0.001 0.005 0.010

0.22 0.17 0.22 0.25 0.27 0.24 0.13 0.21 0.22 0.58

0.06 0.05 0.04 0.04 0.05 0.06 0.02 0.05 0.07

15.5 16.9 14.7 18.4 68.6 15.0 18.0 20.6 112 161

0.7 1.2 1.2 1.8 5.6 1.6 2.2 1.7 73 32

0.008 0.007 0.006 0.008 0.013 0.007 0.008 0.007 0.012 0.015

0.002 0.001 0.001 0.001 – 0.001 0.001 0.001 0.002

NuRD complex

CDK2AP1 CHD4 GATAD2A GATAD2B MBD2 MBD3 MTA1 MTA2 MTA3

2.63 – 0.56 0.62

59.5 – 0.09 0.11

– – 8.8 1.7

– 0.024 0.016 0.014

– 0.011 0.001 0.001

2.9

41.6 24.9

3.1 4.1

0.013 0.011 0.010 0.012

0.001 0.004 0.001 0.002

45.9 – 79.6 53.6 81.5 46.9 45.5 119 48.3

2.5 – 3.1 4.3

0.09 0.11 0.23 0.07

– 0.41 0.38 0.41 0.66 0.33 0.28 0.56 0.28

– – 0.09 0.10

0.68 0.53 0.60 0.41

– 64.1 42.9 33.0 41.9 38.2

3.9 6.8 34 7.9

0.020 0.057 0.020 0.032 0.023 0.020 0.021 0.028 0.019

0.004 0.002

BRMS1 SAP18 SAP30 SAP30L SIN3A SIN3B

0.41d – 0.09 0.20 0.10 0.16

– – 0.01

0.043

– – 0.17 0.17 0.14 0.26

– – 0.04

> 600

0.036 0.027

– – 0.002 – 0.009 0.016

0.03 0.14

> 600 > 600

– – – – – –

0.022 – 0.025 0.049 0.025 0.024

– >20,000 – – – – 0.003 10,485 3,911 0.015 13,236 4,056 0.006 12,671 4,281 15,690 2,739

NCoR complex

GPS2 NCOR1 NCOR2 TBL1X TBL1XR1

0.31 0.25 0.35 0.29 0.27

0.11 0.07 0.02 0.13 0.09

2.9 1.8 1.9 2.6 2.3

0.1 – 0.1 0.1 0.1

0.016 0.017 0.017 0.018 0.014

0.001 0.002 0.002 0.002 0.001

0.20 0.26 0.21 0.26 0.27

0.07 0.06 0.06 0.05 0.04

6.6 5.7 6.8 4.9 4.9

0.5 0.7 1.2 0.3 0.3

0.027 0.024 0.026 0.029 0.025

0.005 0.000 0.001 0.002 0.003

6,068 3,598 4614 4,673 2,956

4,825 365 2,471 1,160 951

DNTTIP1

DNTTIP1 MIDEAS

0.22 0.31

0.07 0.05

55.5 162

4.4 78.2

0.014 0.013

0.001 0.002

0.18 0.18

0.03 0.09

119 156

20 28

0.014 0.013

0.002 0.003

1,187 1,110

286 371

ELM-SANT domain proteins

MIER1 MIER2 MIER3 RERE

0.24 0.59 0.11 0.49

0.03 0.22 0.01 0.12

21.4 26.3 6.3 222

3.7 – – 48.2

0.023

0.003 – 0.001 0.005

0.25 – 0.12 0.32

0.05 0.04 0.20

36.1 37.0 – 272

3.9 – – 57

0.016 – 0.007 0.025

0.002 – – 0.009

1,423 – 954 4,336

892 – 181 1,131

> 30 2.14 1.90 > 30 3.90

– 0.26 0.07 – 0.57

– – – – –

0.66 2.72 > 10 0.06 > 10

0.06 0.40 – 0.01 –

– – – – –

>1 0.96 >1 >1 >1

– 0.02 – – –

Sin3 complex

Selected ALDH1A2 HDACi off-targets BZW2 CBR1 ISOC2 PPP3CA

0.01 0.05

> > > > > >

> > > > >

300 300 300 300 300 300

300 300 300 300 300

– – – – – –

– – – – –

0.029

0.009 0.039 > > > > >

0.1 0.1 0.1 0.1 0.1

0.04 0.11 0.26 0.04

> 600

> > > > >

600 600 600 600 600

0.002 0.023 0.002

– –

594 568 493 748 509 516 536 534 1251 8,23

158 174 95 202 – 130 172 123 327

2,897 1,219 1,429 1,153 1,483 1,223 1,048 1,099

8641 231 212 249 618 140 180

10,528 5,352 >20,000 – >20,000 – >20,000 – >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).

258

VOLUME 29  NUMBER 3  MARCH 2011  nature biotechnology

a

b

Baits:

SIN3

1.0

1.0

Relative enrichment in Sin3A in IP

0.5 ident. in IP: HDAC (class I)

H D H AC D 1 AC LS 2 M D1 T SI A3 N D 3A N TR TT I C ER P1 D F EH YL 1 H MT D 2 A TB C L1 3 XR 1

Enrichment in IP vs ctrl IgG HDAC1 HDAC2 HDAC3 CoREST GSE1 HMG20A HMG20B LSD1 PHF21A RCOR1 RCOR2 RCOR3 ZMYM2 ZMYM3 ZNF217 NuRD CDK2AP1 CHD3 CHD4 GATAD2A GATAD2B MBD2 MBD3 MTA1 MTA2 MTA3 NuRD/Sin3 RBBP4 (shared) RBBP7 Sin3 ARID4A ARID4B BRMS1 BRMS1L* ING1* ING2* SAP30 SAP30L SAP130 SIN3A SIN3B SUDS3 ELM-SANT MIER1 MIER2 MIER3 RERE TRERF1 MIDEAS DNTTIP1 DNTTIP1 HATs CDYL Histone EHMT1 methylation EHMT2 WDR5 WIZ NCoR GPS2 NCOR1 NCOR2 TBL1X TBL1XR1 Histones HIST1H1C HIST1H4A HIST2H2BA

c

1.0

Enrichment IP

29

0.8

9 15

0.6

1

0.4 2

0.2

5

4

0

6

7

–0.2

3

CoREST

Relative enrichment in TBL1XR1 in IP

1.0

NCoR

7

4 3

5

0.8

6

0.6 0.4 0.2 13 28 1 14 16 12 915 17

0 –0.2

1 HDAC1 2 HDAC2 3 HDAC3 4 TBL1XR1 (bait) 5 NCOR1 6 TBL1X 7 GPS2 8 MTA3 (bait) 9 CDK2AP1 10 CHD3 11 CHD4 12 GATAD2A 13 GATAD2B 14 MBD2 15 MBD3 16 MTA1 17 = MTA2

NuRD

–0.4 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1.0 Relative enrichment in MTA3 in IP

19 11 9 20 1 6 15 28 5 22 21 1 14 26 161730 2 7 24 4 37

29

26 19 7 5 27 15 17212022 1 28 11 372 14 16 6 9 39 45 402413 41 388 4 42 23 18 432544 12 10 31 30

32 34 33 3 36

13

35 7

27

HDAC1

0.5

1 HDAC1 2 HDAC2 3 LSD1 (bait) 4 HMG20A 5 RCOR3 6 RCOR1 7 GSE1 8 SIN3A (bait) 9 SAP30 10 ARID4B 11 SAP30 12 SUDS3A 13 BRMS1 14 RBBP4 15 RBBP7

11 12 13 8 14 10

–0.4 –0.4 –0.2 0 0.2 0.4 0.6 0.8 Relative enrichment in LSD1 in IP

31

18

23

HDAC2

0

1

1.0 1

9

11

1

29

4 2

HDAC3

0

7

Enrichment IP

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 sequenceto-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).

14

22 15 28 21 19 2 18 16 17

0

1 24

1 23

26

Protein HDAC1 HDAC2 HDAC3 GSE1 HMG20A HMG20B LSD1 PHF21A RCOR1 RCOR2 RCOR3 ZMYM2 ZMYM3 CDK2AP1 CHD4 GATAD2A GATAD2B MBD2 MBD3 MTA1 MTA2 MTA3

Group HDAC (class I) CoREST

NuRD

No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

Protein BRMS1 SAP30 SAP30L SIN3A SIN3B RBBP4 RBBP7 DNTTIP1 MIDEAS GPS2 NCOR1 NCOR2 TBL1X TBL1XR1 MIER1 MIER2 MIER3 RERE CDYL EHMT1 EHMT2 RREB1 WIZ

Group Sin3

No. 23 24 25 26 27 NuRD/Sin3 28 29 DNTTIP1 30 31 NCoR 32 33 34 35 36 ELM-SANT 37 38 39 40 Other 41 42 43 44 45

35 25 28

36

3

1

33 32

20

5

29 6

LSD1

0.5

MTA3

0

1

1.0 Enrichment IP

© 2011 Nature America, Inc. All rights reserved.

Articles

SIN3A

0

1

1

41

30

30

1

2

37

43

42 45

28

31

nature biotechnology  VOLUME 29  NUMBER 3  MARCH 2011

0

38

1

HDAC1 and HDAC2 (Supplementary Fig. 4 and Supplementary Data Set 3). The quantiDNTTIP1 0.5 fication of relative protein amounts sequestered 0 Relative affinity by the probe matrix upon sequential incubations enabled us to determine apparent dissociation constants (Kdapp) from the IC50 ­values. The resulting deviation between Kdapp 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 prolonged 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

7

TBL1XR1

0

41

37 2 1 2

TRERF1 1

0

Relative affinity

CDYL 1

0

Relative affinity

EHMT2 1

0

Relative affinity

1

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 components 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 stoichiometric binding of one inhibitor molecule per enzyme ­molecule), with 259

260

a

HeLa extract

Probe-matrix bound

Vehicle Aph. Noc. Aphidicolin Nocodazole

Vehicle DNTTIP1 HDAC1

RFU b 3,500

Nocodazole Nocodazole, TSA 10 µM Vehicle Vehicle, TSA 10 µM

3,000 2,500 2,000

*

1,500 1,000 500

3A N SI

(N 3 TA M

(S

uR

in

D

3)

)

) ES T oR (C 1

1 IP TT N

LS D

(M

tro

iD

lI

AC

gG

)

0

D

Deconvolution of HDAC target complexes To differentiate between novel direct targets and novel components of HDAC complexes, we carried out a series of co-immunoprecipitations (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 components 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 analysis, such that specifically co-immunopurified proteins were clearly discriminated from background and differences in complex compositions were directly quantified. Notably, the display of the SIN3Aversus 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 inhibitor 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 confidence 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 inhibitors 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 experimental and a corresponding randomized data set enabled determination 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 proteins 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 copurified 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 expression of DNTTIP1 and HDAC1 in HeLa cells arrested in ­mitosis or in early S-phase. No differences in expression were detected (Fig. 4a).

on

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 Kdapp 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 Kdapp 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 compound set, there were marked differences in the Sin3 inhibition profiles, 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 existence of a distinct mitotic deacetylase complex (MiDAC).

C

© 2011 Nature America, Inc. All rights reserved.

Articles

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.

VOLUME 29  NUMBER 3  MARCH 2011  nature biotechnology

Articles 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 immunoprecipitated 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 divisionarrested cells (Fig. 4b).

b

c

A SA H A PC I-2 Ta 47 ce 81 d R ina om li i ne PC dep I-3 sin Bu 40 fe 51 x Va am a lp ro c at e

TS

VC

SAHA

Cmpd

4

4

Tacedinaline

H2A 1 × ac

3 × ac H2B 1 × ac

PCI−24781

3 × ac 4 × ac H3

1 × ac 2 × ac

H4

0

K me STGGK ac APR

2

0

1 × ac 2 × ac 3 × ac 4 × ac

Histones 3.725

GK ac GGK ac GLGK ac GGAK ac R 4

2

0

VC TS A

Avg. log2 FC vs. unmod 0.000

Fold increase (log2 scale)

PCI−34051

2

2 × ac

2 × ac

Romidepsin

K ac STGGK ac APR

S PC AH Ta -2 A ce 47 d 81 R ina om lin id e PC eps I-3 in Bu 40 fe 51 xa Va ma lp c ro at e

Overlay

Fold increase (log2 scale)

Acetylated tubulin

Fold increase (log2 scale)

Acetylated histone H3

Vehicle

a

Valproate

© 2011 Nature America, Inc. All rights reserved.

Cell-based profiling of HDAC inhibitors To investigate the correlation of proteomics target profiles with substrate selectivity, a subset of the reference inhibitors was subjected 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 inhibitors (­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 acetylation 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 mapping of histone acetylation and methylation marks using quantitative high-resolution MS/MS33 (Fig. 5b and Supplementary Data Set 6). The results confirmed the range of activities observed in chemoproteomics 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).

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).

nature biotechnology  VOLUME 29  NUMBER 3  MARCH 2011

261

Articles

60 40 20 0

OH

O

80

NH O

Bufexamac

60 40 20 0

80

b

O H2N

O

100

HN

1.2

HN

H2N

N

60

Residual binding

80

100

Percent inhibition HDAC3

100

Percent inhibition HDAC6

Percent inhibition HDAC2

a

Cpd AA-1 Cpd AA-2

40 20

1.0 0.8 0.6

HDAC1 (NI) HDAC2 (NI) HDAC3 app (KD = 341 µM) HDAC6 app (KD = 10.7 µM) app HDAC8 (KD = 235 µM) app HDAC10 (KD = 12.3 µM)

0.4 0.2

0

0 40

60

80 100

Percent inhibiton HDAC1 Acetylated histone H3K9

Acetylated tubulin

20

40

60

80 100

0

Percent inhibition HDAC1 Overlay

20

40

60

80 100

1

10 100 1,000 [Bufexamac] (µM)

Percent inhibition HDAC1 100

d

50

100

0 0.01

Percent inhibition / viability

AA-2 Bufexamac

© 2011 Nature America, Inc. All rights reserved.

Control

c

0

0.1 1.0 10 [Cpd AA-2] (µM)

100

100 IC50 = 2.9 µM 50

ac-H4K5 ac-H3K9/K14 ac-tubulin Viability

IFN-α secretion [% inhibition]

20

Percent inhibition / viability

0

IC50 = 8.9 µM

50

0 0.1

1.0 10 [Bufexamac] (µM)

100

0 0.01

0.1 1.0 10 [Bufexamac] (µM)

100

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 (EC 50 = 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).

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 inhibitor To 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 displaying 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 action 34, preferentially affected HDAC6. Bufexamac was subjected to quantitative proteomics profiling 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 262

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). DISCUSSION Gene 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 VOLUME 29  NUMBER 3  MARCH 2011  nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

Articles activity of compounds is assessed without use of recombinant purified 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 targets 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 binding to the hydroxamate matrix displayed minor differences across a panel of cell lines and tissues. However, two proteins lacking an obvious 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 bidirectional hierarchical clustering of the proteomics target profiles of 16 inhibitors (Fig. 2b) guides (i) the classification of drugs in selectivity 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 chemotypes represented by hydroxamates and aminobenzamides, with several 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 controlled 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 nature biotechnology  VOLUME 29  NUMBER 3  MARCH 2011

c­ ellular 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 antibodies 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 components 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 isochorismatase domain protein ISOC2, and several other Zn2+-dependent metalloenzymes. In the protein dimension, the clustering data delineate target protein 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 HDACcontaining 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 sensitive 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 function 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 specialized 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 component 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 263

© 2011 Nature America, Inc. All rights reserved.

Articles strong ­associations with other proteins in the inhibition profiles were detected. HDAC6 has recently been implicated in chromatin regulation44 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 identified 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 anti­inflammatory 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 confirms 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. Methods Methods 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. Acknowledgments This 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 CONTRIBUTIONS A.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 INTERESTS The 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).

264

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).

VOLUME 29  NUMBER 3  MARCH 2011  nature biotechnology

Articles 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).

© 2011 Nature America, Inc. All rights reserved.

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).

nature biotechnology  VOLUME 29  NUMBER 3  MARCH 2011

265

© 2011 Nature America, Inc. All rights reserved.

ONLINE METHODS

Reagents. 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 (H3AcK9/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 synthesized 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 cultured 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.

c­ ompound 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 contained 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% inhibition, 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 quantitative 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 different antibodies directed against the same (or different) antigen(s) and two ‘mock IP’ samples. Enzymatic deacetylation assays. The enzyme activity of purified recombinant HDAC1, HDAC2, HDAC3-NCoR and HDAC6 (1 µg of protein per well) was measured using the Active Motif HDAC Assay Kit in 96-well format 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 concentration) 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.

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 equilibrated 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-overend 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.

Sample preparation for MS. Gels were cut into slices across the entire separation range and subjected to in-gel digestion21. For acquisition of doseresponse 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.

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

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

nature biotechnology

doi:10.1038/nbt.1759

© 2011 Nature America, Inc. All rights reserved.

nano-LC system (Eksigent 1D+) coupled to LTQ-Orbitrap mass spectrometers (Thermo-Finnigan). Peptides were separated on custom 50 cm × 75 µM (internal 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 combined 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 precursors 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 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-to­interference measure29. The heat maps in Supplementary Figure 2a–d are based on the accumulated reporter ion responses for each protein divided by its molecular weight (10 percentile bins). This enables accurate relative quantification between different conditions for the same protein and similarly to spectrum count based methods 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:

doi:10.1038/nbt.1759

FA(i) =

A(i)

n

∑ j =1 A( j)

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 acquisition for a subset of proteins29. To compare selectivities of compounds displaying different absolute potencies relative potencies were calculated as (pIC50 – min(pIC50))/(max(pIC50) – min(pIC50)) for each experiment. Apparent dissociation constants (Kdapp) 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 previously described. IC50 values for 21 proteins were calculated for each experiment 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 experiments were normalized to values between 0 and 1 (5,683 unique proteins). The average normalized pIC50 (anpIC50) value was calculated across all compounds included for each protein. The anpIC50 of each protein is linked to the normalized enrichment (nPD) values determined in each immunoprecipitation 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 resulting 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 databases 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 quantified proteins identified with at least 9 independent experiments were considered. 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 identified 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

nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

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 quantified 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 enrichment as previously described50 and separated using SDS gel electrophoresis. For indirect immunofluorescence analysis, HeLa cells were plated to subconfluency on polylysine-coated glass chamber slides and treated after recovery

nature biotechnology

with the indicated compounds for 4 h. Samples were fixed in cold methanol, permeabilized with Triton-X100, blocked in 1% BSA and treated for immunodetection of acetylated tubulin and acetylated histone H3. Cells were counterstained for nucleic acids using 4′,6-diamidino-2-phenylindole (DAPI). Multichannel fluorescence microscopy was performed on an Olympus IX70 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 chargeenhancement 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).

doi:10.1038/nbt.1759