Design and synthesis of benzodiazepine analogs as isoform-selective human lysine deacetylase inhibitors

Article abstract of DOI:10.1016/j.ejmech.2016.12.032

A comprehensive investigation was performed to identify new benzodiazepine (BZD) derivatives as potent and selective human lysine deacetylase inhibitors (hKDACis). A total of 108 BZD compounds were designed, synthesized and from that 104 compounds were biologically evaluated against human lysine deacetylases (hKDACs) 1, 3 and 8 (class I) and 6 (class IIb). The most active compounds showed mid-nanomolar potencies against hKDACs 1, 3 and 6 and micromolar activity against hKDAC8, while a promising compound (6q) showed selectivity towards hKDAC3 among the different enzyme isoforms. An hKDAC6 homology model, refined by molecular dynamics simulation was generated, and molecular docking studies performed to rationalize the dominant ligand-residue interactions as well as to define structure-activity-relationships. Experimental results confirmed the usefulness of the benzodiazepine moiety as capping group when pursuing hKDAC isoform-selectivity inhibition, suggesting its continued use when designing new hKDACis.

Full text of DOI:10.1016/j.ejmech.2016.12.032

Accepted Manuscript
Design and synthesis of benzodiazepine analogs as isoform-selective human lysine
deacetylase inhibitors
D. Rajasekhar Reddy, Flavio Ballante, Nancy J. Zhou, Garland R. Marshall
PII:
S0223-5234(16)31038-8
10.1016/j.ejmech.2016.12.032
EJMECH 9123
DOI:
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 8 November 2016
Revised Date: 30 November 2016
Accepted Date: 16 December 2016
Please cite this article as: D.R. Reddy, F. Ballante, N.J. Zhou, G.R. Marshall, Design and synthesis of
benzodiazepine analogs as isoform-selective human lysine deacetylase inhibitors, European Journal of
Medicinal Chemistry (2017), doi: 10.1016/j.ejmech.2016.12.032.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT
Design and Synthesis of Benzodiazepine Analogs as Isoform-Selective Human Lysine
Deacetylase Inhibitors
*
D. Rajasekhar Reddy, Flavio Ballante, Nancy J. Zhou, and Garland R. Marshall
Department of Biochemistry and Molecular Biophysics, Washington University School of
Medicine, St. Louis, MO 63110
*
Corresponding author E-mail: garlandm@gmail.com (Garland R. Marshall)
ABSTRACT
A comprehensive investigation was performed to identify new benzodiazepine (BZD) derivatives
as potent and selective human lysine deacetylase inhibitors (hKDACis). A total of 108 BZD
compounds were designed, synthesized and from that 104 compounds were biologically
evaluated against human lysine deacetylases (hKDACs) 1, 3 and 8 (class I) and 6 (class IIb). The
most active compounds showed mid-nanomolar potencies against hKDACs 1, 3 and 6 and
micromolar activity against hKDAC8, while a promising compound (6q) showed selectivity
towards hKDAC3 among the different enzyme isoforms. An hKDAC6 homology model, refined
by molecular dynamics simulation was generated, and molecular docking studies performed to
rationalize the dominant ligand-residue interactions as well as to define structure-activity-
relationships. Experimental results confirmed the usefulness of the benzodiazepine moiety as
capping group when pursuing hKDAC isoform-selectivity inhibition, suggesting its continued
use when designing new hKDACis.

ACCEPTED MANUSCRIPT
KEYWORDS
Epigenetics; Histone Deacetylase (HDAC); Lysine Deacetylase (KDAC); Benzodiazepine
BZD); Structure-based drug design (SBDD)
(
INTRODUCTION
Being an epigenetic target, lysine deacetylases (KDACs) generally referred to as histone
deacetylases (HDACs), are a family of enzymes that are responsible for deacetylation of lysine
residues in both histone and non-histone substrates [1], emerged as an important target in the
development of potential therapeutic agents [2-6]. These enzymes are involved in a wide variety
biological processes, including cell differentiation, angiogenesis, and apoptosis [3, 7, 8]. The 18
isoforms of KDAC are classified into four distinct groups: class I (KDACs 1, 2, 3, and 8), class
II (class IIa (KDACs 4, 5, 7, and 9), class IIb (KDACs 6 and 10)), class III (sirtuins) and class IV
(
KDAC11) based on their size, number of catalytically active sites, subcellular locations, and
sequence homology with respect to their yeast counterparts [9-12]. Except for class III KDACs
+
(
sirtuins) that are NAD -dependent proteins, the other enzymes are all zinc dependent [9, 13, 14].
2
+
Numerous structurally diverse KDACs inhibitors (KDACis) against Zn -dependent KDACs
have been explored [15, 16], and some of them FDA approved while others are at various stages
of clinical evaluation [17, 18]. Romidepsin (FK-228) [19, 20], vorinostat (SAHA) [21],
belinostat (PXD101) [22], and panobinostat (LBH589) [23, 24], (Figure 1) have been approved
by the U.S. Food and Drug Administration (FDA) for the treatment of cutaneous T-cell
lymphoma (CTCL), peripheral T-cell lymphoma (PTCL), and multiple myeloma, respectively.
Chidamide, a KDACi developed in China, has been recently approved by the China Food and

ACCEPTED MANUSCRIPT
Drug Administration for the treatment of PTCL [25]. However, most KDACis are class I
selective (FK-228, PXD101) or pan-KDAC inhibitors (SAHA, LBH589). These nonselective or
partially selective KDAC inhibitors usually lead to undesirable biological side effects, such as
fatigue, nausea, and cardiotoxicity [26-30]. To avoid the side effects, an increasing number of
investigations are focusing on the development of isotype-selective KDAC inhibitors [14, 17]. In
continuation of these efforts, benzodiazepines (BZDs) have also been used as cap groups to
discover isoform-selective KDAC inhibitors (Figure 2) [31, 32]. During the course of our
investigation, an article appeared that also utilized BZD as a cap group (Figure 2) [33, 34].
Figure 1. Examples of potent KDAC inhibitors.

ACCEPTED MANUSCRIPT
Herein, we report the design of novel C3-substituted benzodiazepines with an amino acid side
2
+
chain as cap group and at N1 linear alkyl chain bearing different functions as the Zn -chelating
group (Figure 2) targeting human lysine deacetylase (hKDAC) isoforms based on analysis of
structure-activity relationships (SARs); a library of BZD analogs has been synthesized. The lead
compounds from this library exhibit potent class I KDAC selectivity.
Figure 2. KDAC inhibitors with BZD scaffold as cap group.
RESULTS
Chemistry
C3-substituted BZDs (2a-2u) were synthesized from readily available 2-aminobenzophenone in
a three-step process, often without the need for purification until the cyclization step. N1-
Substituted BZD analogs (3a-3s) were synthesized by treating BZDs (2a-2s) with sodium
hydride and alkylation with methyl-5-bromovalerate in DMF at room temperature. BZD acids
(4a-4s) were obtained by the hydrolysis of BZD-methyl esters (3a-3s) with 1M KOH aqueous
solution in MeOH or by using LiOH in MeOH/ H O (9:1). BZD hydroxamic acids (5a-5s) were
2
generated with a mixed anhydride approach with ethyl chloroformate (ECF) and reacting with
.
NH OH HCl (Scheme 1).
2

ACCEPTED MANUSCRIPT
Finally, the BZD hydroxamates with free side chain were obtained by cleaving the protecting
groups. BZD hydroxamates 6g, 6m, 6r, 6s, 6n, and 6p were obtained by cleaving the protecting
groups on 5g, 5m, 5r, 5s, 5n, and 5p compounds side chain with TFA. Side chain protecting
groups cleavage with TFA/ TIS on compounds 5j, 5k, 5o, and 5q provided hydroxamate BZDs
6
j, 6k, 6o, and 6q. Compound 6h was obtained from 5h by employing the
trifluoromethanesulfonic acid and debenzylation of 5i with excess of BF .OEt and EtSH
afforded 6i (Scheme 2).
3
2

ACCEPTED MANUSCRIPT
BZD compounds (8r-8u) were synthesized by mixed anhydride formation from their acids (7r-
7
u), which were obtained by tert butyl deprotection of 2r-2u compounds with TFA. BZD
hydroxamates (11r, 11s) were obtained from BZD acids 7r, 7s by mixed anhydride approach and

ACCEPTED MANUSCRIPT
reacting with methyl-4-aminobutanoate HCl salt [35]. The resulted esters 9r, 9s on hydrolysis
provided acids 10r, 10s. These acids are converted to hydroxamic acids 11r, 11s by reacting with
hydroxylamine HCl salt (Scheme 3). BZDs 13r, 13s were synthesized by mixed anhydride
formation from their acids (12r, 12s), which were obtained by tert butyl deprotection of
compounds 4r, 4s with TFA (Scheme 4).
Bioactivity Against Isolated hKDAC Isoforms
All the synthesized BZD derivatives were assayed (for details, see the Experimental Section) for
their inhibitory activity against recombinant hKDACs 1, 3 and 8 (Class I KDACs) and hKDAC 6

ACCEPTED MANUSCRIPT
(Class II KDACs). In vitro assays were performed using the Caliper EZ Reader II system
(Perkin-Elmer Caliper Life Sciences, USA). The candidates were firstly tested in duplicate at 30
µM (see Supplementary material Tables S3-S4 for all the BZDs inhibitor % values over the 4
considered hKDACs). Those compounds showing an inhibitory activity ≥50% at 30 µM were
then assayed using two replicates in a 10-point dose curve with 3-fold serial dilution starting
from 30 µM, for which IC values as well as dose-response curves were then derived (Tables 1
50
and 2 and Supplementary material Table S5 for dose-response curves). As positive controls ten
well-known hKDACis were used (Table 3, Supplementary material Table S6 for dose-response
curves).
Table 1. C3-substituted BZD analogs hKDAC IC (µM) values.
5
0
b
Biological response
c
IC (µM)
a
50
Entry
ID
8s
R
1
hKDAC isoform
1
3
6
8
1
16.92 >30 >30 >30
2
3
11r
11s
17.38 3.21 1.76 >30
22.87 5.17 7.16 >30
a
Only compounds showing at least one detectable IC are reported (see Supplementary material
5
0
b
Table S3 for the complete list). Compounds were tested in a 10-point dose with 3-fold serial
dilution starting from 30 µM. IC values and dose-response curves were derived for compounds
c
5
0
with inhibition ≥50% at 30 µM. See Supplementary material Tables S3 and S5 for all inhibition
data and dose-response curves.

ACCEPTED MANUSCRIPT
Table 2. N1, C3-Substituted BZD analogs hKDAC IC50 (µM) values.
b
Biological response
IC50 (µM)
hKDAC isoform
c
a
Entry
ID
R1
1
3
6
8
1
2
5a
5b
H
Me
3.84
12.67
0.66
1.80
1.89
2.95
6.72
24.75
3
4
5c
5.27
5.90
0.60
0.57
1.35
2.33
19.83
>30
5d
5
6
5e
5f
10.96
6.43
1.03
0.43
3.75
1.70
>30
20.28
7
8
5g
5h
9.00
7.99
1.21
0.75
5.05
5.37
>30
>30
9
5i
4.55
0.91
2.05
29.54
1
0
5j
>30
4.31
6.27
>30
1
1
1
2
5k
5l
9.24
8.51
0.83
0.49
0.99
1.98
23.48
12.67
1
3
5m
6.18
0.85
2.87
3.30

ACCEPTED MANUSCRIPT
1
4
5n
>30
5.21
6.92
8.39
1
1
5
6
5o
5p
>30
2.55
0.59
4.74
3.06
>30
4.63
21.79
1
7
5q
>30
4.29
16.17
>30
1
1
8
9
5r
5s
15.7
4.57
4.15
1.03
6.43
2.35
22.39
26.07
2
2
0
1
6g
6h
>30
4.94
0.18
6.35
0.58
>30
0.46
8.11
2
2
2
2
3
4
6i
6j
11.23
22.53
13.95
1.80
3.10
1.94
11.58
5.38
1.29
>30
>30
>30
6k
2
2
5
6
6m
6n
>30
5.29
1.01
8.46
1.95
>30
>30
16.84
2
2
7
8
6o
6p
13.88
6.54
1.32
0.95
4.98
>30
>30
12.12

ACCEPTED MANUSCRIPT
2
9
6q
1.44
0.22
2.63
>30
3
3
3
3
0
1
2
3
6r
6s
>30
>30
12.60
8.71
19.84
16.00
>30
>30
>30
>30
8.52
13r
13s
9.99
14.97
10.39
2.57
1.74
a
Only compounds showing at least one detectable IC are reported (see Supplementary material
5
0
b
Table S3 for the complete list). Compounds were tested in a 10-point dose with 3-fold serial
dilution starting from 30 µM. IC values and dose-response curves were derived for compounds
c
5
0
with inhibition ≥50% at 30 µM. See Supplementary material Tables S4 and S5 for all inhibition
data and dose-response curves.
Table 3. Standard compounds (positive controls) hKDAC IC (µM) values.
5
0
a
Biological response
b
Entry
ID
IC50 (µM)
KDAC isoform
1
3
6
8
1
2
Entinostat (MS-275) [36]
1.48 [35]
2.33 [35]
0.79 [35]
1.36 [35]
>30 [35]
9.29 [35]
>30 [35]
>30 [35]
c
Largazole [37]
Panobinostat (LBH589) [23,
3
0.02
0.06
0.13
1.78
24]
4
5
6
7
8
9
PCI-34051 [38]
SD-L-256 [39]
Trichostatin A (TSA) [40]
Tubastatin A [41]
T247 [42]
14.41 [35]
3.48 [35]
0.015 [35]
2.87 [35]
1.11 [35]
>30 [35]
0.47 [35]
0.020 [35]
0.77 [35]
3.94 [35]
4.57 [35]
1.61 [35]
0.038 [35] 4.55 [35]
0.014 [35] 2.34 [35]
0.49 [35]
>30 [35]
>30 [35]
>30 [35]
Vorinostat (SAHA) [43]
4 [44]
0.0070 [35] 0.0014 [35] 0.0014 [35] 0.50 [35]
27.37 20 >30 1.09
10
a
Compounds were tested in a 10-point dose with 3-fold serial dilution starting from 30 µM.
b
IC values and dose-response curves were derived for compounds with inhibition ≥ 50% at 30
5
0
c
µM. The thioester form of largazole was assayed as previously reported [35].
See Supplementary material Table S6 for dose-response curves.

ACCEPTED MANUSCRIPT
Inhibitor Potency. Among the BZD compounds, the highest inhibitor activities were generally
detected towards the hKDAC3 isoform (Tables 1 and 2): 6h (a N1, C3-substituted BZD) showed
the highest potency against hKDAC3 (IC = 180 nM), with a mid-nanomolar activity against
5
0
hKDACs 1 and 6 (IC = 460 nM and 580 nM, respectively) and a low micromolar inhibition
5
0
against hKDAC8 (IC = 8.11 µM). Compound 6h was the only BZD endowed with an IC
50
5
0
value lower than 1 µM against hKDAC1 and hKDAC6 (as reported above), together with 5k (a
N1, C3-substituted BZD having an IC50 = 990 nM vs hKDAC6). Similarly as 6h, compound 6q
(a N1, C3-substituted BZD) showed an IC50 value of 220 nM against hKDAC3, but a lower
activity (thus higher selectivity) when considering the other 3 isoforms (IC50 values of 1.44 µM,
2
.63 µM and >30µM against hKDAC1, 6 and 8, respectively). Other 11 BZD compounds (5f, 5l,
d, 5p, 5c, 5a, 5h, 5k, 5m, 5i and 6p; all N1, C3-substituted BZDs) were detected to be most
5
active against the hKDAC3 isoform showing an IC50 ~1µM. No BZD derivatives showed
interesting inhibitor activity against hKDAC8: 5m (IC = 3.3 µM) was detected as the most
5
0
active.
Isoform Selectivity. By comparing the 104-benzodiazepine derivatives with their relative
isoform-related activities (Tables 1-2 and Supplementary material Tables S3-S4), it was
possible to rationalize a series of features required for inhibitory potency and isoform selectivity.
hKDAC1. 28 BZDs showed an inhibitory percentage value greater than 50% at 30 µM (with a
range of IC values comprised between 460 nM and 22.87 µM; only compound 6h showed an
5
0
IC < 1µM). Among them, only three C3-subsituted BZDs (8s, 11r, 11s) were detected, but
5
0
showing lower inhibitory activities (IC50 values of 16.92 µM, 17.38 µM and 22.87 µM,
respectively). All of the derivatives with an IC <30µM were hydroxamic acids. Conversely,
5
0
those compounds without inhibitory activity had no hydroxamic moiety as a zinc-binding group,

ACCEPTED MANUSCRIPT
confirming the effectiveness of this moiety for ligand-protein recognition: for example,
compounds 5a and 4a (both are N1, C3-substituted, with a hydrogen in C3, and 4 carbon atoms
as linker in N1) differ only by the presence of the hydroxamic moiety (characterizing 5a), but
show a completely different hKDAC1 inhibition activity (IC50 = 3.84 µM and >30µM,
respectively).
About the C3-substituted derivatives, the C3 configuration affected the biological response; the
L-derivatives were more active than those compounds with a D-configuration. For example,
compound 8s (a C3-substituted BZD), having a (L)-C3, has a higher activity (IC50 = 16.92 µM)
than its enantiomer (compound 8u, IC50 >30 µM), thus suggesting a primary role of the C3
configuration in BZD-hKDAC1 recognition.
hKDAC3. Compared to hKDAC1, a higher number of tested compounds (a total of 35) showed
an inhibition percentage greater than 50% at 30 µM (IC values between 180 nM and 12.6 µM).
5
0
Of those, thirteen showed an IC50 ~1µM. Noteworthy, the highest inhibition potencies were
obtained against this enzyme isoform (e.g. compounds 6h and 6q as reported above). Among
those derivatives showing an IC50 <30 µM, only two are C3-substituted (11r and 11s, IC50 values
of 3.21 µM and 5.17 µM, respectively); interestingly, even those with a 6- or 7-atom linkers
showed higher activities compared to hKDAC1 (IC values equal to 17.38 µM and 22.87 µM,
50
respectively). This suggests that hKDAC3 accepts ligands characterized by a longer linker.
Similar to hKDAC1, all of the derivatives with IC <30 µM are hydroxamic acids, but the D/L
50
configuration at C3 (when considering C3-substituted derivatives), gave an opposite effect on
hKDAC3 inhibition compared to the hKDAC1 case: for example, compound 8u (D-derivative)
showed a higher inhibition percentage (29.45% at 30 µM) with respect to its L-enantiomer
(4.57% at 30 µM, see Supplementary material Table S3).

ACCEPTED MANUSCRIPT
hKDAC6. 34 BZD compounds showed an inhibition percentage greater than 50% at 30 µM (with
a range of IC values comprised between 580 nM and 19.84 µM); only two from this series (6h
5
0
and 5k), had an IC ≤1µM (580 nM and 990 nM respectively). As for hKDAC1 and 3, among
5
0
the most active derivatives (those with an IC ≤30 µM), only two (11r and 11s, IC values equal
5
0
50
to 1.76 µM and 7.16 µM, respectively) are C3-substituted, while all are hydroxamic acids. Thus,
similar to hKDAC3, ligands characterized by a longer linker are well accepted by hKDAC1. The
effect of the D/L-configuration at C3 (for C3-substituted derivatives) is similar to the case of
hKDAC1; the L-configuration increased the inhibition potency (i.e. 8s, an L-derivative showing
an inhibition of 36.51% at 30µM was more active than its enantiomer 8u which has
approximatively half the potency, 19.02% at 30 µM, see Supplementary material Table S3).
hKDAC8. Compared to the other isoforms, hKDAC8 was generally less inhibited by the tested
compounds. In fact, only 14 BZDs resulted with an inhibition percentage value greater than 50%
at 30 µM (showing a range of IC values comprised between 3.3 µM and 29.5 µM), and none of
5
0
these with an IC ≤1 µM. Similarly to the results for the other isoforms, the most active
5
0
compounds (with an IC50 ≤30 µM) were hydroxamic acids, but (conversely from the other
targeted enzymes) none of them were C3-substituted derivatives. The first four most active
compounds are 5m, 5a, 6h and 5n (IC50 = 3.30 µM, 6.72 µM, 8.11 µM and 8.39 µM,
respectively), characterized by Tyr(tBu), Gly, Cys and Trp(Boc) amino acid side chains suggest
that aromatic rings with terminal aliphatic chains (as in 5m and 5n) or short aliphatic/polar
moieties (as in 5a and 6h) are preferred (e.g. compare 5m with 5l; 5n with 6n). To the contrary,
the low activity of 6q and 6p (characterized by an Arg and Lys) indicates the negative impact of
a positively charged group on hKDAC8 inhibition.

ACCEPTED MANUSCRIPT
Isoform Selectivity. Beside the inhibitory potency, isoform selectivity is an important issue
during the discovery of new KDACis. Given the number of enzyme isoforms herein, we propose
to summarize the selectivity of an inhibitor using a global isoform-selectivity index (GISIndex)
for each specific hKDAC isoform. To obtain the GISIndex, all the partial selectivity indexes
towards each investigated isoform were derived (SIs, Supplementary material Table S7), and the
average computed. When considering hKDACs 1, 3, 6 and 8, for each enzyme isoform, three
partial selectivity indexes (SIs) can be derived. As an example, when considering hKDAC1,
three SIs can be computed: from hKDAC3 (denoted as 1/3), hKDAC6 (denoted as 1/6), and
hKDAC8 (denoted as 1/8); then the hKDAC1 GISIndex is derived by averaging the three partial
SIs (Supplementary material Table S7). As a result, a selective KDACi will show a higher
GISIndex coefficient only towards one hKDAC isoform. Table 4 reports the positive
GISIndexes characterizing the tested compounds (BZDs and standards) while those with a
GISIndex equal to zero are omitted (see Supplementary material Table S7 for the complete table
of results). Tubastatin A was the most selective compound, showing one relevant GISIndex
(142.38) towards hKDAC6, while compounds characterized by at least two similar GISIndexes
among the enzyme isoforms, were not as selective (e.g. TSA and SAHA, as expected since they
are characterized as pan-inhibitors). Among the BZD derivatives, compound 6q was the most
selective with a detectable preference for hKDAC3 (GISIndex = 51.62, and negligible values
from the other isoforms), followed by a series of BZDs: 5d, 5f, 5h, 6p, 6h, 5p, 5l and 6n with a
hKDAC3 GISIndex greater than 15 while negligible for other isoforms (Table 4). All of the 9
compounds composing the above series, are N1, C3-substituted BZD analogs and hydroxamic
acids; moreover, even if characterized by different amino acid chains, their activity range against

ACCEPTED MANUSCRIPT
hKDAC3 varies by only a factor of 5.61 (being 6h the most active and 6n the least active,
showing an IC = 180 nM and 1.01 µM, respectively).
5
0
Table 4. hKDACs 1, 3, 6 and 8 global isoform selectivity indexes. The tested compounds
have been ranked for each isoform according to the relative global isoform selectivity index
value. Standard compounds are reported as italic characters.
hKDAC1
GIS Index
02.40
hKDAC3
GIS Index
haKDAC6
haKDAC8
a
a
ID
TSA
ID
GIS Index
ID
GIS Index
ID
1
3
2
1
1
120.71
75.83
51.62
25.32
24.89
22.36
22.02
19.27
17.07
16.94
16.66
15.76
15.46
14.69
14.47
13.24
13.14
12.34
12.13
10.68
10.61
9.78
9.63
7.84
7.55
6.93
6.29
5.92
5.65
5.08
4.92
4.84
4.64
4.05
SAHA
142.38
120.71
39.91
11.86
11.02
9.58
8.01
6.93
6.20
5.54
5.24
5.14
4.99
4.66
4.56
4.35
4.23
4.22
3.80
3.64
3.57
3.25
3.24
3.19
3.15
2.94
2.88
2.57
2.46
2.36
2.36
1.97
1.86
1.85
1.25
1.24
1.10
1.08
1.01
0.86
0.83
Tubastatin A
33.32
23.66
1.19
0.62
0.59
0.41
PCI-34051
2.83
3.81
9.20
3.51
LBH589
SAHA
TSA
6q
SAHA
4
5n
5m
TSA
6k
5k
11r
T247
MS-275
SD-L-256
MS-275
13s
7
6
.55
.30
.62
.87
.35
.16
.15
.90
.77
.69
.57
.25
.25
.23
.11
.05
.91
.85
.72
.72
.69
.65
.59
.58
.58
.50
.48
.44
.44
.37
6q
6h
5d
5f
5h
6p
6h
5p
5l
6n
5c
5e
5i
5k
6o
5g
5s
Tubastatin A
6n
5
2
2
2
2
1
1
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Largazole
SD-L-256
SD-L-256
5c
5i
5f
5d
13s
6h
13r
5i
6p
5s
8s
5d
5p
5c
5h
6i
5g
5f
5e
5k
6o
6k
LBH589
5s
5b
5o
6q
5e
5l
6j
LBH589
6i
Largazole
PCI-34051
5o
6k
5b
5a
5q
6j
T247
11r
5m
5j
6g
6m
11s
5r
13s
6s
5j
6g
6o
5p
5g
11s
6m
5h
5r
5a
PCI-34051
5b
6n
5a
11r
5l
5r
6j
11s
4
5n
6s
5q
3.78
3
3
3
2
2
1
1
1
.41
.06
.05
.30
.26
.92
.92
.59
5m
Largazole
6r
Tubastatin A
6i
6p
13r
5n
6r
a
Global Isoform Selectivity Index (GISIndex): is the average value of all the partial selectivity
indexes when considering a certain hKDAC isoform (see Supplementary material Table S7).

ACCEPTED MANUSCRIPT
1
Among these 9 compounds, even with a polar or charged R group (cysteine or arginine) had the
highest hKDAC3 inhibition potency (as 6h and 6q showing IC50 = 180 nM and 220 nM
respectively), the presence of a hydrophobic, or aromatic amino acid still guaranteed a decent
activity (e.g. 5l or 6n which are characterized by a phenylalanine or tryptophan, showed IC =
5
0
4
90 nM and 1.01 µM respectively), suggesting a high degree of active-site flexibility by
hKDAC3. Conversely, the discriminatory effect of the different R groups was dramatically
1
higher when considering hKDAC1 and hKDAC6 among this series of 9 compounds. In fact, the
hKDAC1 inhibitor potencies drastically decrease when a polar amino acid (like a cysteine or
arginine characterizing 6h and 6q respectively) was replaced with a hydrophobic one (a leucine
as in 5d), even more if an aromatic ring was present (phenylalanine or tryptophan in 5l and 6n
respectively). Interestingly, 6p (characterized by a lysine at R ) was a weaker hKDAC1 inhibitor
1
compared to 6q (an arginine derivative), thus suggesting the positive effect of having a
guanidinium group (as in 6q respect that one in 6p). Conversely, hKDAC6 inhibition was not
much affected with bulky groups present at R (tryptophan or phenylalanine in 6n and 5l
1
respectively). Lastly, the weak hKDAC8 inhibition from this series of 9 compounds suggests a
lack of specific chemical interactions for enzyme inhibition; a decrease of inhibition effect is
observed when considering a short and polar amino acid group (cysteine in 6h, IC = 8.11 µM)
5
0
was replaced by a longer and charged/polar one (i.e. 6q and 6p, characterized by an arginine or
lysine, respectively with both showing an IC >30 µM).
5
0

ACCEPTED MANUSCRIPT
Two compounds, 6k and 11r, were more selective towards hKDAC6 (Table 4), while 5k having
similar GISIndexes values for hKDAC3 and 6 (13.14 and 11.02, respectively) cannot be
considered as selective. Interestingly, 6k, 11r and 5k all featured an amide group characterizing
the cap portion, suggesting a potential role of this moiety when targeting hKDAC6 selectivity.
To further clarify the discriminatory BZDs-hKDACs interaction as well as those necessary for
inhibition potency, molecular modeling studies were performed.
Molecular Modeling
To rationalize the biochemical results and the isoform selectivity, a comparative structure
analysis was performed among the 4 selected hKDAC isoforms also used as targets for
molecular docking of the synthesized 104 BZD derivatives (Tables 1 and 2, Supplementary
material Tables S3-S4) as well as standard compounds (positive controls, Table 3). Crystal
structures of hKDAC1, hKDAC3, and hKDAC8 (PDB codes 4BKX, 4A69, 3RQD, respectively)
were prepared as previously reported [45], while a hKDAC6 homology model was derived and
refined with molecular dynamics simulation.
hKDAC6 model generation
The homology model (a hKDAC6-SAHA complex) was generated and subsequently submitted
to MD simulation for 95 ns to study the protein’s structural flexibility and to obtain a refined
model to be used during docking studies (see Molecular Modeling section in Supplementary
material). SAHA was included in the active site during model generation, due to its high
inhibitory activity against the hKDAC6 (IC = 1.4 nM, Table 3) and to prevent collapse of the
5
0
active site, thus providing a reasonable model structure to be used for future docking. Multiple
crystallographic structures of human and zebrafish KDAC6 catalytic domains (hCD1 and hCD2,

ACCEPTED MANUSCRIPT
and zCD1 and zCD2, respectively), released [46, 47] when our study was completed, provided a
template for validation of our hKDAC6 homology model (see Supplementary material Figure
S3). By superimposing our KDAC6 homology model with the relative X-ray protein structure
(PDB code: 5EDU) [46] an RMSD of 1.05Å was detected, suggesting a general good
approximation of the predicted model’s backbone. As expected, major conformational
differences were detected for residues characterizing an important loop for molecular recognition
of ligands endowed with high conformational variability (see Supplementary material Figure
S3C), as a reasonable result of the different bound ligand (SAHA in the homology model and
TSA in the X-ray complex), and experimentally supported by the different bioactivity of the two
hydroxamic acid derivatives (27-fold IC difference, see Table 3).
5
0
hKDACs 1, 3, 6 and 8 structural comparison.
By superimposing the four hKDAC isoforms (1, 3, 6 and 8, Figure 3A), one can highlight
sequence conservation percentages, i.e., for those amino acid portions (close to the active site)
characterized by the highest variability and involved in enzyme selectivity (Figure 3B). Five
sequences (1 to 5, Figure 3C; see Supplementary material Figure S4) were selected and
compared (Figure 4).
Sequence 1. The characterizing loop (Figure 4, Supplementary material Figure S4) has a similar
residue sequence when hKDAC1 and 3 are considered (Gln26-Gly27-His28-Pro29-Met30-Lys31
versus Ala20-Gly21-His22-Pro23-Met24-Lys25, respectively), while more differences were
detected in hKDAC6 (Ser498-His499-His500-Pro501-Glu502-Val503). The highest
dissimilarities are detected when considering hKDAC8 characterized by lysine (Lys33) and
isoleucine (Ile34) residues, recognized to cause conformational distortion of the binding pocket
[48], thus suggesting a role of this sequence in enzyme adaptation.

ACCEPTED MANUSCRIPT
Sequence 2. Located next to Sequence 1 (Figure 3C), this characterizing loop (Figure 4,
Supplementary material Figure S4) has a different primary sequence when considering the four
hKDAC isoforms. hKDAC1, 3 and 8 have the most similar amino acid sequence, Gly-X-Asp-
Cys, where X is a non-conserved residue represented by a glutamate, aspartate and tyrosine
(Glu98, Asp92 and Tyr100), respectively. This difference should affect the hKDAC8 selectivity
with respect to the other two isoforms, due to a higher dissimilarity of a tyrosine residue with
respect to a glutamate or aspartate. hKDAC6 shows a higher number of non-conserved residues,
thus suggesting a different behavior when binding a ligand.
Sequence 3. Located in an inner part of the active site (Figure 4, Supplementary material Figure
S4), below the zinc ion, this sequence is characterized by a conserved leucine residue in
hKDAC1 and 3 (Leu139 and Leu133, respectively); while hKDAC6 is characterized by a
glycine (Gly609), suggesting recognition of bulkier groups, as a consequence of a decreased
steric hindrance. Conversely, the presence of a bulky tryptophan (Trp141) in hKDAC8 hinders
access to this portion of the active site channel.

ACCEPTED MANUSCRIPT
Figure 3. Superimposed hKDAC1 (PDB 4BKX), hKDAC3 (PDB 4A69), hKDAC6 (homology
model) and hKDAC8 (PDB 3RQD). The zinc ion is depicted as grey ball. A: hKDAC1 (grey
colored), 3 (green colored), 6 (magenta colored) and 8 (cyan colored); B: hKDAC1, 3, 6 and 8
color-coded according to the percentage of conserved residues (gradient color from blue: 25% to
yellow: 100%), are depicted as ribbons (left side) and surfaces (right side), respectively. C:
highlighted sequences (1 to 5) with lowest residue conservation in the proximity of the active site
(see Supplementary material (Figure S4) for relative sequences).
Sequence 4. By analyzing this characterizing loop among the four different enzyme isoforms
(Figure 4, Supplementary material Figure S4), a pair of phenylalanine residues is conserved in

ACCEPTED MANUSCRIPT
hKDACs 3, 6 and 8 (Phe199 and Phe200; Phe679 and Phe680; Phe207 and Phe208,
respectively); on the contrary, in hKDAC1 one phenylalanine is replaced by a tyrosine residue
(Tyr204). It seems reasonable that the Phe-Phe or Phe-Tyr pairs interact with the linker-cap
portions of a hKDAC inhibitor, thus contributing to stabilizing a compound mainly by
hydrophobic/steric interactions in the linker or cap regions. The presence of a Tyr204 in
hKDAC1 suggests a tendency of hKDAC1 to welcome a more electrostatic interaction (i.e. H-
bonds) with respect to the other 3 KDACs. When considering the other residues of sequence 4:
starting from those adjacent to the Phe-Phe or Phe-Tyr pair, a proline (Pro206, Pro201, Pro681
and Pro209, in hKDAC 1, 3, 6 and 8 respectively) is conserved, while a glutamate (Glu203),
threonine (Thr203), threonine (Thr678) and a glycine (Gly206) characterize the four isoforms (1,
3
, 6 and 8 respectively), suggesting a role of these residues in ligand-protein recognition.
Differences among other side chains in the loop characterizing this sequence could play a role
for conformational adaptation when interacting with a ligand. Another peculiarity of hKDAC6 is
the higher number of residues due to an insertion compared to the other isoforms (compare with
Supplementary material Figure S4), thus suggesting further flexibility of the active site when
binding a ligand.
Sequence 5. The characterizing side chains of this segment (Figure 4) do not necessarily point
towards the active site, but are potentially involved as the size of the ligand increases. The loop
is composed by a conserved residue sequence (Arg-Leu-Gly-Cys-Phe) in hKDAC1 and 3, while
Pro-Leu-Gly-Gly-Cys in hKDAC6, and Pro-Met-Cys-Ser-Phe in hKDAC8 occur (see
Supplementary material Figure S4), thus reflecting a different contribution of charged and polar
side chains which could determine ligand selectivity.

ACCEPTED MANUSCRIPT
hKDAC1
hKDAC3
hKDAC6
hKDAC8
Sequence 1
Sequence 2
Sequence 3
Sequence 4
Sequence 5
Figure 4. Side chain view of lower conserved sequences close to the active site (1 to 5, Figure
C): hKDAC1 (grey), 3 (green), 6 (magenta) and 8 (cyan). The zinc ion is depicted as grey ball.
3

ACCEPTED MANUSCRIPT
Molecular Docking
To rationalize the structure activity relationship of the synthesized 104 BZD derivatives,
molecular docking calculations were applied using the four hKDAC isoforms (1, 3, 6 and 8) as
target structures.
Docking Assessment. As previously reported [45], a comprehensive docking assessment study
has been performed by us (using all the co-crystal hKDAC structures available at the time) to
detect the most reliable docking protocol when using the mammalian lysine deacetylates as
target. As suggested by the docking assessment results [45], Surflex-Dock [49] was selected to
predict the binding poses of the benzodiazepine compounds when the best-docked (BD) [45]
poses were considered. A further validation has been performed assuming that every active
BZD-hKDACi must interact with the zinc ion (thus having the zinc binding group close to the
catalytic ion), conversely for the non-active compounds. Thus, all the best-docked poses
obtained from docking the BZD derivatives (Tables 1 and 2) were analyzed by correlating the
distances between the closest inhibitor’s zinc-binding group (ZBG) atom and the zinc ion with
the relative biological responses (Figure 5). As depicted in Figure 5, the docking protocol
discriminates the active compounds from the non-active ones among the four enzyme isoforms.
In fact, all of the most active BZDs (showing lower IC (µM) values) were predicted with the
5
0
zinc-binding group close to the zinc ion (≈ 2 Å). While for the least active ones (characterized by
higher IC50 (µM) or lower inhibition percentage values), the closest inhibitor’s atom was
progressively farther from the catalytic zinc. The data further confirmed the reliability of the
adopted protocol, and the absence of any predicted false negative when considering the BZD
derivatives (no BD poses of compounds with an appreciable inhibitory capability were predicted
with the ZBG far from the zinc ion).

ACCEPTED MANUSCRIPT
KDAC1
KDAC3
KDAC6
KDAC8
Figure 5. Distances between the zinc ion and the closest BZDs ZBG atoms (y axes) vs the
relative biological responses (x axes).

ACCEPTED MANUSCRIPT
In the case of the hydroxamic acid derivatives, the characteristic bidentate chelation with the zinc
ion was generally missed by the docking predictions (see below), as a consequence (from our
point of view) of limited protein flexibility and dynamics during the docking simulation, and
further magnified by the presence of the large benzodiazepine capping group.
Binding Mode Analysis. As reported above, BZD compound 6h showed, towards hKDAC1, 3, 6,
but not 8, the maximum inhibitory potency among the 104 tested BZDs. Thus, the best docked
(BD) [45] poses (Figure 6) from each enzyme isoform were investigated to detect those
interactions predicted to be necessary for protein recognition when a BZD was present. Beside
the importance of a hydroxamic moiety as zinc binding group, a common interaction, between
the hydrogen of a zinc-coordinated histidine (His141 in hKDAC1, His172 in hKDAC3, His651
in hKDAC6 and His180 in hKDAC8) and the 2-oxo group of the BZD scaffold was generally
detected (Figure 6), and judged relevant for the inhibitor potency. As a consequence, a certain
length of the linker group (4 carbon atoms generally gave the best results) was necessary to
establish a correct 2-oxo group/histidine H-bond interaction. The low selectivity of 6h among
hKDAC1, 3, 6 suggests that the benzodiazepine moiety with a short R group (i.e. a cysteine as
1
pendant amino acid), is indiscriminately well accepted by these series of enzymes and capable to
establish a packed network of steric/hydrophobic interactions (mostly with hKDAC3), and steric
ones as in hKDAC1 (with Phe205), and hKDAC6 (with Pro748) (see Figure 6). On the contrary,
6
h showed low inhibition potency against hKDAC8: by analyzing its docked pose, no particular
electrostatic interactions as well as hydrophobic ones were detected, when compared with the
other enzyme isoforms. As reported above 6q showed a similar potency, compared to 6h, against
hKDAC3 and its docked pose suggests the establishment of an H-bond interaction between its
pendant amino acid in R (arginine) and the backbone carbonyl oxygen of Tyr198. Since
1

ACCEPTED MANUSCRIPT
molecular docking is generally performed using a fixed protein conformation, a direct
involvement of the Tyr198 hydroxyl group is plausible with a consequent conformational
rearrangement of the characterizing loop (see also sequence 4 in Figure 4). Conversely as 6h, 6q
showed a lower potency against hKDAC1: the predicted docked pose shows an opposite binding
conformation of 6q with respect to 6h (compare Figure 7 with Figure 6), while an H-bond
interaction between the arginine pendant amino acid in R and the backbone’s carbonyl group of
1
Glu203 was detected (Figure 7). Similarly as reported for hKDAC3, a direct involvement of the
Glu203 carboxylate group with a consequent remodeling of the loop (see also sequence 4 in
Figure 4) is not excluded. The binding pose of 6q in hKDAC6 is quite similar to the one
predicted for 6h only when considering the benzodiazepine ring, while its pendant amine group
(arginine) interacts with the backbone’s carbonyl group of a threonine (Thr678, Figure 7; see
also sequence 4 in Figure 4). 6h (compare Figure 6 and 8) suggests a detrimental effect of
having positive charged groups interacting with sequence 4 (Figure 4). When considering
hKDAC8, the predicted conformation of the benzodiazepine moiety characterizing 6q overlaps
that from 6h, while its arginine guanidinium group interacts with the backbone’s carbonyl group
of Gly206; this result suggests as anticipated when comparing the different enzyme isoforms (see
hKDACs 1, 3, 6 and 8 structures comparisons), the roles of Glu203 (hKDAC1), Tyr198
(hKDAC3), Thr678 (hKDAC6) and Gly206 (hKDAC8) in ligand/protein recognition. Binding
mode analysis for those BZDs having the best selectivity by analyzing the GISIndex (Table 4),
was also performed. Among the first series of compounds showing the highest GISIndex towards
hKDAC3 (6q, 5d, 5f, 5h, 6p, 6h, 5p, 5l and 6n, see Isoform Selectivity paragraph), the first five
compound’s docked poses were further investigated (Figure 8, 6q was already analyzed above).
5
1
d, 5f, 5h are all characterized by hydrophobic pendant amino acids in R (a leucine, a

ACCEPTED MANUSCRIPT
methionine and a tert-butyl cysteine, respectively), while 6p has a charged side chain (lysine).
Docking calculations predicted all of this compounds bound similarly to hKDAC3 with the
different amino acids substituents placed in a region (between Phe199, and Leu266, Figure 8)
which accepted, preferably, hydrophobic moieties. It should be remarked that an H-bond
interaction with Tyr198 seems to guarantee the highest hKDAC3 potency and selectivity (as the
case of 6q). The other series of compounds (6k, 11r), with a detectable selectivity against
hKDAC6, are showed in Figure 9 together with 5k (which conversely has a poor selectivity
between hKDAC3 and 6, Table 4). 6k and 11r are N1, C3-substituted and C3-substituted
derivatives, respectively, and characterized by a similar biological activity against hKDAC6.
Their docked poses (Figure 9) suggest that the benzodiazepine scaffold establishes
hydrophobic/aromatic interactions with Phe566, Gly619, Phe620 and Leu749, while maintaining
an H-bond interaction between the 2-oxo group and His651. These interactions appear required
for a decent inhibitory activity (in fact the presence of a glutamine as in 6k doesn’t dramatically
affect the biological response). 5k (an N1, C3-subtituted BZD) which differs for the presence of
2 2 1
a –(CH ) -CO-NH-Trt as group in R shows a similar inhibition potency towards hKDAC3 and
6
, confirming the capability for hKDAC6 to accept bulkier moieties. The presence of the
Arg673, Asp675 and Arg709 in hKDAC6 (see 5k in Figure 9) could increase the inhibitor
selectivity between hKDAC3 and hKDAC6. The most important evidence collected by
correlating the BZD chemical structures with the biological responses by molecular modeling are
summarized in a pharmacophore scheme (Figure 10) as a useful starting point for the
development of a next generation of BZD derivatives as hKDACis.