Pharmacophore assessment through 3-D QSAR: Evaluation of the predictive ability on new derivatives by the application on a series of antitubercular agents

Article abstract of DOI:10.1021/ci400132q

Pharmacophoric mapping is a useful procedure to frame, especially when crystallographic receptor structures are unavailable as in ligand-based studies, the hypothetical site of interaction. In this study, 71 pyrrole derivatives active against M. tuberculosis were used to derive through a recent new 3-D QSAR protocol, 3-D QSAutogrid/R, several predictive 3-D QSAR models on compounds aligned by a previously reported pharmacophoric application. A final multiprobe (MP) 3-D QSAR model was then obtained configuring itself as a tool to derive pharmacophoric quantitative models. To stress the applicability of the described models, an external test set of unrelated and newly synthesized series of R-4-amino-3-isoxazolidinone derivatives found to be active at micromolar level against M. tuberculosis was used, and the predicted bioactivities were in good agreement with the experimental values. The 3-D QSAutogrid/R procedure proved to be able to correlate by a single multi-informative scenario the different activity molecular profiles thus confirming its usefulness in the rational drug design approach.

Full text of DOI:10.1021/ci400132q

Subscriber access provided by UNIV OF REGINA
Article
Pharmacophore Assessment Through 3-D QSAR:evaluation
of the predictive ability on new derivatives by the
application on a serie of antitubercularagents.
Rino Ragno, and Flavio Ballante
J. Chem. Inf. Model., Just Accepted Manuscript • DOI: 10.1021/ci400132q • Publication Date (Web): 25 Apr 2013
Downloaded from http://pubs.acs.org on April 28, 2013
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Chemical Information and Modeling is published by the American
Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the
course of their duties.

Page 1 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Pharmacophore Assessment Through 3-D QSAR: evaluation of the predictive ability on new
derivatives by the application on a serie of antitubercular agents.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
§,†
§,‡
,‡
‡
†
Laura Friggeri, Flavio Ballante,*, Rino Ragno,* Ira Musmuca, Daniela De Vita,, Fabrizio
†
†
†
†
Manetti,¤ Mariangela Biava, Luigi Scipione, Roberto Di Santo,+, Roberta Costi,+, Marta
Feroci,^ and Silvano Tortorella.
†
‡
Rome Center for Molecular Design, Dipartimento di Chimica e Tecnologie del Farmaco, Sapienza
Università di Roma, P. le A. Moro 5, 00185 Roma, Italy.
†
Dipartimento di Chimica e Tecnologie del Farmaco, Sapienza Università di Roma, P. le A. Moro
5, 00185 Roma, Italy.
+Istituto Pasteur-Fondazione Cenci Bolognetti, Dipartimento di “Chimica e Tecnologie del
Farmaco”, “Sapienza” Università di Roma, P.le A. Moro 5, 00185 Rome, Italy
¤Dipartimento di Biotecnologie, Chimica e Farmacia, Università degli Studi di Siena, Via Aldo
Moro 2, I-53100 Siena, Italy.
^ Dipartimento di Scienze di Base e Applicate per l’Ingegneria, Sapienza University of Rome, Via
Castro Laurenziano 7, I-00161 Rome, Italy.
§
L.F. and F.B. contributed equally to this work
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 2 of 37
1
2
3
4
Abstract
Pharmacophoric mapping is a useful procedure to frame, especially when crystallographic receptor
structures are unavailable as in ligand-based studies, the hypothetical site of interaction. In this
study, 71 pyrrole derivatives active against M. tuberculosis were used to derive through a recent
new 3-D QSAR protocol, 3-D QSAutogrid/R, several predictive 3-D QSAR models on compounds
aligned by a previously reported pharmacophoric application. A final multi probe (MP) 3-D QSAR
model was then obtained configuring itself as a tool to derive pharmacophoric quantitative models.
To stress the applicability of the described models, an external test set of unrelated and newly
synthesized series of R-4-amino-3-isoxazolidinone derivatives found to be active at micromolar
level against M. tuberculosis, was used and the predicted bioactivities were in good agreement with
the experimental values. The 3-D QSAutogrid/R procedure proved to be able to correlate by a single
multi-informative scenario the different activity molecular profiles thus confirming its usefulness in
the rational drug design approach.
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

Page 3 of 37
Journal of Chemical Information and Modeling
1
2
3
1. Introduction
4
5
6
7
8
Tuberculosis (TB), an infectious disease mainly caused by Mycobacterium tuberculosis (Mtb),
remains a major public health problem and causes ill-health among millions of people each year.
TB ranks as the second leading cause of death from an infectious disease worldwide, after the
human immunodeficiency virus (HIV-1). The 2012WHO Global tuberculosis report estimates there
are almost 9 million new cases and 1.4 million TB deaths.1 Moreover, two billion people are
estimated to be latently infected with Mtb, and the 10% of them reactivating to active TB with
major risk relative to immigrants from endemic areas, people with HIV-1 infection, individuals with
underlying diseases (silicosis, diabetes mellitus, malignant conditions).2
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Currently, the standard treatment comprises: first line drugs, such as isoniazid (INH), pyrazinamide
(PZA), ethambutol (EMB), rifampin (RIF); and second line drugs, such as ethionamide (ETH), p-
aminosalicylic acid (PAS), capreomycin, amynoglicosides, D-cycloserine (DCS) and
fluoroquinolones.3
The required long-term drug treatment, due to the high persistence of Mtb, combined with poor
compliance of the patients, highly contributes to develop drug resistant strains, particularly
multidrug-resistant (MDR, resistant at least to INH and RIF) and extensively drug-resistant (XDR,
resistant at least to INH, RIF and three of second line class of anti-TB drugs). Recently a more
dangerous form of bacilli, named totally drug-resistant (TDR) showing in vitro resistance to all
first- and second –line drugs tested have been isolated.4, 5
To reduce this increasing problem, antitubercular drugs are used with specific therapeutic protocols
under direct observation therapy short course (DOTS) conditions.6
The need for new shorter therapeutic regimens and new classes of drugs active on MDR, XDR and
TDR MTB drives pharmaceutical research to accelerate in the development process of new anti-TB
drugs.7 Continuing our research on anti-TB agents,8, 9 here we report the assessment of a previously
reported pharmacophore model10 through 3-D QSAutogrid/R, a recent introduced quantitative
ligand-based design protocol.11 The developed 3-D QSAR models were tested for their predictive
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 4 of 37
1
2
3
4
5
6
7
8
9
ability on a series of new independently synthetized R-4-amino-3-isoxazolidinone derivatives 1a-e,
2a-f and 3h-i (Table 1). These compounds have been designed to evaluate the effects on
antitubercular activity due by the introduction of acyl substituents on N(2) atom of oxoisoxazolidine
ring and on amino group.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 1. R-4-amino-3-isoxazolidinone derivatives: monocarbamates (1a-e), dicarbamates (2a-f)
and amides (3h,i).
H
R1
O
H
R1
O
O
O
N
N
R2
O
O
H
O
H
R2
N
N
O
O
O
Monocarbamates 1a-e
Dicarbamates 2a-f
#
R1
R2
H
#
R1
R2
1a
2a
1b
1c
H
H
2b
2c
2d
2e
1d
1e
H
H
2f
H
H
N
O
F3C
O
H
HN
N
O
N
O
O
O
F3C
Cl
Cl
3h
3i
Amides 3h, 3i
ACS Paragon Plus Environment

Page 5 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
2 Results and Discussion
2.1 Ligand-Based Design
6
7
8
9
A first pharmacophore model for anti-TB activity was previously developed by us12 using a series
of 32 imidazole derivatives with interesting antitubercular activity, adopting the HipHop13 method.
The final model was then optimized14 and finally characterized by four pharmacophoric features as
follows: an hydrogen bond acceptor feature (HBA), two aromatic ring features (RA1, RA2), and an
hydrophobic feature (HY); and applied recently to different antimycobacterial agents.10, 15 Even if
this model is able to describe the needed structural properties for antitubercular activity and identify
the possible antimycobacterial candidates within large molecular databases, it doesn’t permit to
correlate quantitatively biological activity of the compounds with their structural features. This
limitation is due to the fact that the model was obtained by application of the qualitative approach
referred to as the common feature hypothesis generation method. In addition, as for the specific
case of the newly synthesized monocarbamates (1a-e), dicarbamates (2a-f) and amides (3h-i) of R-
4-amino-3-isoxazolidinone (Figure 1, discussion in the External Test Set Prediction Analysis
paragraph), a proper evaluation may be difficult when a partial overlap of the investigated
compounds with the defined pharmacophore areas is established. In this perspective the use of a
three-dimensional quantitative approach is useful and several 3-D QSAR PLS models, characterized
by a training set (Table 2, Table S1 for numeric reference) of 71 published antitubercular agents,10,
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
15-19
were built through the 3-D QSAutogrid/R11 protocol: 8 mono-probe (see Table S5 for probes’
definition) 3-D QSAR PLS models were generated and optimized via the CAPP11 procedure (Table
3 and 4) and a final multi probe (MPGRS) model (Tables 5 and S5) was then derived to correlate
the pharmacophoric features required for antitubercular activity with molecular structures. Activity
data, originally determined as MIC (g/mL) values, were transformed to pMIC values on molar
basis.
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 6 of 37
1
2
3
4
Table 2. Structure and antimycobacterial activity against M. tuberculosis 103471 of the pyrrole
5
6
derivatives used as training set for the generation of the 3-D QSAR models
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
compda
Rb
B
R1
R2
R3
pMICc
4.68
5
1
2
3
4
5
6
7
8
2-F-Ph
2-Cl-Ph
2-Cl-Ph
2-F-Ph
2-F-Ph
2-Cl-Ph
2-Cl-Ph
2-F-Ph
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
2-F-Ph
A
B
2-F-Ph
2-F-Ph
4.09
5
A
B
2-Cl-Ph
2-Cl-Ph
2-Cl-Ph
2-Cl-Ph
-naphthyl
-naphthyl
-naphthyl
4.4
A
B
5.02
4.41
4.11
A
9
B
A
B
2-F-Ph
2-Cl-Ph
2-Cl-Ph
CH3
CH3
CH3
4.11
4.13
4.13
10
11
-naphthyl
2-Cl-Ph
Ph
12
13
14
15
16
17
18
19
20
21
A
B
B
A
B
A
B
A
A
B
-naphthyl
4-F-Ph
Ph
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
4.13
4.36
4.36
5.30
4.47
5.58
5.30
5.60
4.66
4.06
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-Cl-Ph
4-Cl-Ph
Ph
4-Cl-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
2-F-Ph
2-F-Ph
Ph
ACS Paragon Plus Environment

Page 7 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
22
23
24
25
26
27
28
A
B
A
B
B
A
B
Ph
Ph
CH3
CH3
CH3
CH3
CH3
CH3
CH3
2-F-Ph
2-F-Ph
Ph
4.96
4.36
4.38
4.07
4.07
4.1
2-Cl-Ph
2-Cl-Ph
Ph
Ph
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
2-Cl-Ph
Ph
-naphthyl
-naphthyl
Ph
Ph
4.09
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
A
B
B
A
B
B
A
B
A
B
A
B
A
B
B
A
A
B
A
B
A
B
A
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
-naphthyl
-naphthyl
Ph
4.10
4.09
4.33
5.00
4.70
4.08
5.98
4.37
4.98
4.07
5.34
5.00
5.30
5.00
4.68
5.58
4.98
4.40
4.68
4.10
5.64
5.03
5.30
Ph
Ph
4-F-Ph
2-Cl-Ph
2-Cl-Ph
2-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-CH3-Ph
3-CH3-Ph
2-CH3-Ph
2-CH3-Ph
2,4-Cl2-Ph
2,4-F2-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
2-Cl-Ph
2-Cl-Ph
2-F-Ph
4-F-Ph
4-F-Ph
4-CH3-Ph
3-CH3-Ph
3-CH3-Ph
2-CH3-Ph
2-CH3-Ph
2,4-Cl2-Ph
2,4-Cl2-Ph
2,4-F2-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
4-F-Ph
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 8 of 37
1
2
3
4
5
6
7
8
9
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
B
A
A
A
A
A
A
A
A
A
A
A
A
B
A
A
A
A
A
A
2,4-F2-Ph
4-F-Ph
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
CH3
C2H5
C2H5
C2H5
C2H5
C2H5
C2H5
C2H5
4-F-Ph
4-C2H5-Ph
4-i-propyl-Ph
4-F-Ph
4.40
5.60
6.21
5.30
5.61
5.90
6.22
6.23
6.53
5.90
5.91
6.23
6.23
5.33
6.20
5.26
5.58
5.28
5.30
5.00
4-F-Ph
4-C2H5-Ph
4-C3H7-Ph
4-Cl-Ph
4-Cl-Ph
4-Cl-Ph
4-Cl-Ph
4-CH3-Ph
4-C2H5-Ph
4-C3H7-Ph
4-i-propyl-Ph
4-Cl-Ph
4-F-Ph
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
4-F-Ph
4-CH3-Ph
4-C2H5-Ph
4-C3H7-Ph
4-i-propyl-Ph
4-Cl-Ph
4-Cl-Ph
4-Cl-Ph
4-Cl-Ph
4-Cl-Ph
4-CH3-Ph
Ph
Ph
Ph
4-F-Ph
4-F-Ph
Ph
2-F-Ph
4-F-Ph
2-F-Ph
2-F-Ph
c.Compound enumeration was assigned on the basis of the original increasing numbering from the oldest to the most
recent reference. Table S1 shows the connections between the new and original enumerations.
aA = thiomorpholin-4-yl and B = 4-methylpiperazin-1-yl
bpMIC = -Log [MIC(µM) x 10-6]
ACS Paragon Plus Environment

Page 9 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
A
B
C
Figure 1. A: monocarbamates (1a-e); B: dicarbamates (2a-f); C: amides (3h-i) of D-4-amino-3-
isoxazolidinone placed in the reported pharmacophoric model:14 HY (hydrophobic feature), RA
(aromatic feature), HBA (hydrogen bond acceptor feature).The four pharmacophoric features are
color-code according the original reference.
59
60
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 10 of 37
1
2
3
4
5
6
7
8
Three of the best mono-probe 3-D QSAR models, A, HD and NA (Table 4 and Figure S1),
accounting for different interaction patterns, were selected for further analysis and the relative 3-D
plots were inspected (Figure 2, Figures S2-S4). A comparison between these plots and the original
pharmacophoric model14 was performed to check for spatial superposition of plot regions and
pharmacophoric features (compare Figures 2A-2C with Figure 2D). Interpretation of the PLS-
coefficients plots could be helped considering the Equation 1.0 where Cn is the CPLS coefficient in
the nth grid point, Xn is the actual field in the nth grid point, Y the biological activity and n the
number of grid points.
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Y = C1X1+C2X2+…+CnXn+e.
Equation 1. General equation for QSARs
As addressed by Equation 1, CPLS coefficients provide both interpretation of training set data
(explaining the relative influence of each grid point by means of size and sign) and prediction of
test set molecules’ biological activity, Y ; an interaction characterized by a positive (repulsive) field
Xn into a region with positive PLS-coefficient Cn will produce a positive effect (Cn × Xn product is
positive), denoting a positive influence on Y (higher pMIC); the opposite is valid if the field or the
PLS-coefficient have negative Xn or Cn, respectively. A positive effect (still considering pMICs
activities) could be produced as well by a negative (attractive) field Xn into a region with negative
PLS-coefficient Cn (-Cn × -Xn product is positive) and the opposite effect if the field or the
PLS_coefficient have positive Xn or Cn, respectively.
As a result, four areas of the PLS-coefficients plots were distinguished over the N1, C2, C3 and C5
substituents of the pyrrole ring which overlap the pharmacophoric features HY, RA1, HBA and
RA2, respectively, thus suggesting a good agreement between QSAR and pharmacophoric models.
All mono-probe 3-D QSAR models clearly suggest that the presence of bulky groups as substituent
at N1, C2, C5 is preferred, especially in N1 and C5 (HY and RA2 features). Moreover, the PLS-
coefficient plots within thiomorpholines and methylpiperazinyl derivatives, (i.e. compounds 60 and
ACS Paragon Plus Environment

Page 11 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
21 in Figure 2) and even more both PLS-coefficients and activity contribution plots (Figures S2-S4)
show that at the C3 position steric features are required within certain limits. In particular,
concerning the HBA feature, the HD model clearly shows that attractive interactions involving the
sulfur of the thiomorpholine group of 60 (the most active compound, Figure S3A), increase the
biological activity, while if these are missing or replaced by repulsive interactions, as for 21 (the
least active compound), the biological effect decreases (Figure S3B). Therefore, a bulky group like
thiomorpholinomethyl, also able to participate in electrostatic interactions, such as hydrogen bonds,
is preferred at the C3 position. As for the structural features required for activity, the simultaneous
analysis of PLS-loading and score plots were very useful to carry out the most relevant variables
from the models (loading plots) and interpret the patterns seen in the score plots. Interesting is the
case of the A probe model: starting from the first principal component (PC1), the presence of two
clusters, differing each other for their conformational properties, is clearly showed in the scores plot
(Figure S5A). As shown in Figure S6 by superimposing the most influencing compounds (absolute
higher score values) for each cluster to the PLS-loadings, the most important molecular feature in
the PC1 space is related to both different spatial orientations and conformations. Light grey
molecules, that are molecules in the positive field cluster (positive PC1), had a higher activity and
fill the lower part of the region between the RA2 and HBA features but not between RA1 and HBA.
On the contrary, molecules with lower pMICs fill the area between the RA1 and HBA. Therefore a
given derivatives able to preferably occupy the region between the RA2 and HBA features, should
be endowed of a higher activity than a molecule filling the area between the RA1 and HBA. PC2
and PC3 respectively gave information about substituents at C3 (Figures S5B, S7 and S8),
suggesting the presence of bulky groups in the upper areas between the RA2 and HBA and over the
HBA features have a detrimental effect on the biological activity.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 3. CAPP settings
Min Value
Parameter
PCO
Max Value
10
Step
1.0
0
0
Zeroing
0.05
0.005
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 12 of 37
1
2
3
4
5
0
MSDCO
5
1
PCO: Positive Cut Off, Zeroing: zeroing of very low data points, MSDCO: Minimum SD Cut Off.
6
7
8
9
Table 4. 3-D QSAutogrid/R PLS models statistical results (CAPP process was applied)
model
P
A
C
HD
NA
N
OA
e
PC
3
3
3
3
3
3
4
4
r2
q2
q2
r2
q2
V
LOO
K5FCV
YS
YS
0.92
0.92
0.91
0.91
0.91
0.91
0.88
0.91
0.86
0.86
0.85
0.86
0.85
0.85
0.78
0.85
0.85
0.85
0.84
0.85
0.85
0.85
0.76
0.84
0.36
0.37
0.39
0.31
0.32
0.36
0.40
0.35
-0.33
-0.33
-0.31
-0.33
-0.30
-0.33
-0.48
-0.44
3758
4492
1217
531
477
658
1
2
3
4
5
6
7
8
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
468
4412
d
P: Autogrid Probe, PC: optimal number of principal components/latent variables, r2:
conventional square-correlation coefficient; q2LOO: cross-validation correlation coefficient
using the leave-one-out method; q2K5FCV: cross-validation correlation coefficient using the
k-fold cross-validation with 5 random groups and 100 iterations; r2YS: average square-
correlation coefficient obtained after Y-scrambling process using 100 iterations; q2
:
YS
average cross-validation correlation coefficient using the leave-one-out method obtained
after Y-scrambling process using 100 iterations; V: number of active variables.
ACS Paragon Plus Environment

Page 13 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
A
B
C
D
Figure 2. The most active (60 in blue) and the less active (21 in magenta) compounds are shown. A:
PLS-coefficients contour maps derived from A probe analysis (contour levels: 80%; positive: red,
negative: blue); B: PLS-coefficients contour maps derived from HD probe analysis (contour levels:
85%; positive: red, negative: blue); C: PLS-coefficients contour maps derived from NA probe
analysis (contour levels: 75%; positive: red, negative: blue). D: pharmacophoric features derived
from the original pharmacophoric model:14 HY (hydrophobic feature), RA (aromatic feature), HBA
(hydrogen bond acceptor feature).
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 14 of 37
1
2
3
4
Application of Multi-Probe Guided Region-Variable Selection
5
6
7
8
By application of the Multi Probe Guided Region Selection (MPGRS package), as implemented in
3-D QSAutogrid/R,11 a multi-probe (MP) 3-D QSAR model was derived, representing, the best of
our knowledge, the first quantitative pharmacophoric model able to correlate the structural features
of pyrrole derivatives with their biological data. The optimal MP 3-D QSAR model was
characterized by a PCFL:SL=1:3,11 and as previously reported,11 its associated statistical coefficients
(Table 5, Figure S9) were similar to those of the mono-probe models but the interpretation was
greatly enhanced. Applying a q2 threshold value of 0.4, the most relevant MIFs sub-regions were
selected (Figure 3) to build the multi-probe MIF and the resulting MP model condensing in one all
the suggestions retrieved by the analysis conducted on the mono-probe models. In particular, the
MP PLS-loadings in association with the MP score plots identified the same conformational
differences, addressed by the mono-probe models, as the most discriminating aspect in molecular
clustering, for example: starting from PC1:1 to PC1:2 a similar clustering in the score plots and in
meaning for the descriptors to those in the A mono probe model was noticed (compare Figures S10-
S12 and Figures S5-S7), confirming the above assumptions (effect of difference in spatial
arrangement and conformation). The MP PLS-coefficient plot showed that the most important
regions were spatially and chemically overlapping with the pharmacophoric model14 (compare
Figures 3B and 3D). Taking into account the probe type with the associated PLS-coefficient sign,
bulky groups seemed to be required at the N1, C2 and C5 positions (positive coefficients);
furthermore negative PLS-coefficients were spread in the proximity of these areas and the fact that
chlorine and fluorine substituents are associated to activity enhancement, these additional areas can
be related to some electrostatic molecular environment (Figure 3B and 3C) in agreement with the
pharmacophoric model.16, 18 In addition, regarding C3 position (in the lower part of HBA and
slightly extended towards RA2) the model indicates that a limited steric repulsion is tolerated and
electrostatic endowed groups could be profitable for the activity. Further information about the
HBA feature were derived overlapping the clustered molecules (Figure S10B) with both PLS-
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

Page 15 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
loadings and PLS-coefficients at PC1:3 (Figure 4A and 4B): the implementation of different probes
(such as NA and HD) suggested that the presence of bulky groups in the upper areas between the
RA2 and HBA and over the HBA feature might have a detrimental effect on the biological activity;
i.e. considering the methylpiperazinyl moiety (characterizing most of the negative clustered
molecules, Figure 4B2) the methyl group fits the HD areas characterized by a negative PLS-
coefficient, while the thiomorpholinomethyl moiety (that discriminate the positive clustered
molecules, Figure 4B1) satisfies both steric and electrostatic features leading to higher activities. In
this scenario the MP model was able to increase the resolution of the HBA region revealing an extra
partial steric role.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 5. MPGRS. Multi Probe model statistical results
MPGRS 3-D QSAR
PCFL:SL
r2
q2
q2
SDEP LOO
SDEP K5FCV
r2
q2
YS
LOO
K5FCV
YS
0.88
0.80
0.80
1:3
0.32
0.32
0.31
-0.31
PCFL:SL: optimal number of principal first level (FL) and second level (SL) components for the
MPGRS model; r2: conventional square-correlation coefficient; q2LOO: cross-validation correlation
coefficient using the leave-one-out method; q2K5FCV: cross-validation correlation coefficient using
the k-fold cross-validation with 5 random groups and 100 iterations.
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 16 of 37
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
B
A
D
C
Figure 3. MPGRS. A: key points: the points are color coded according to that reported in Table S5;
B: key points with PLS-coefficients contour maps (contour levels: positive 85%, red; negative 95%,
blue); C:top view, key points with PLS-coefficients solid contour maps (contour levels: positive
85%, red; negative 95%, blue). The most active (60 in blue) and the less active (21 in magenta)
compounds are shown. D: pharmacophoric features derived from the original pharmacophoric
model:14 HY (hydrophobic feature); RA (aromatic feature), HBA (hydrogen bond acceptor feature).
ACS Paragon Plus Environment

Page 17 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
A
B1
B2
B
Figure 4. MPGRS. PLS-loadings contour maps at PC1:3 (contour levels: 75%; positive: orange,
negative: cyan) with PLS-coefficients (mesh levels: positive 85%, red; negative 95%, blue) and key
points (see Table S5 for color coding). The ten most important molecules for each cluster are
plotted and color coded (compounds in the positive loading field in light gray; compounds in the
negative loading field in dark grey). A:side view; B: frontal view; B1: frontal view of only positive
clustered molecules; B2: frontal view of only negative clustered molecules. HY (hydrophobic
feature), RA (aromatic feature), HBA (hydrogen bond acceptor feature).
55
56
57
58
59
60
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 18 of 37
1
2
3
4
External Test Set Prediction Analysis.
5
6
7
8
The eight 3-D QSAutogrid/R mono probe models were externally validated using the 13 newly
synthesized monocarbamates (1a-e), dicarbamates (2a-f) and amides (3h-i) (Table 1). A fact must
be emphasized: since the training set was composed only by pyrrole derivatives to directly compare
the quantitative models with the original pharmacophoric assumptions, the resulting quantitative
structure-activity relationships were based, mostly, on the characteristics of the scaffolds composing
the training set. This may result in a limitation of the models to predict the activities of other
molecular classes: in this case, specifically, a major difference between the two sets, training set and
test set, was represented by the fact that the former was characterized by the pyrrole ring, which
permits a quadruple branching able to satisfy simultaneously the different pharmacophoric areas; on
the contrary, the compounds of the test set were characterized by a double branching. Despite this
fact, and considering also that the test set molecules showed similar activity values (total pMIC
activity range = 1.31 log unit),acceptable errors of prediction (SDEP coefficients all below the unit
except for the d model) were obtained (Table 6); but an analysis focused only on the statistical
SDEP values or experimental vs predicted plots (Figure S13) could be misleading. In fact, analyzing
only the statistical results might seem that the PLS models had good predictive ability towards the
isoxazolidinone derivatives, while considering only the experimental vs predicted plots the same
conclusion could not be reached. In this case it was helpful to analyze both of these information,
together with the average absolute error of predictions (AAEP, Table S6), for each molecule from
all the 3-D QSAR mono probe models and finally their placement in the 3-D space of PLS-
coefficients. Analyzing, for each test set molecule, the AAEP from all the eight mono probe models
stood out the good predictive capacity toward 10 of these, while for 1a, 2c and 2e the AAEPs were
1.05, 1.50 and 1.72 respectively. These compounds were over-predicted and this can be sought
precisely in the dependence of the model from the training set congenerousness and from
consequent inevitable alignment limitations. Furthermore 1a was predicted more active than 1d
likely due to the fact that its isoxazolidinonic carbonyl group was perfectly superimposed to the
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

Page 19 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
training set most active compound (60) thiomorpholinic sulfur atom, showing how important was
for the models the presence of a group capable to accept hydrogen bonds in the HBA space. It
should be stressed, however, that such molecules (1a, 2c and 2e), probably, would have been
discarded by adopting the original pharmacophoric model,14 in fact: compound 1a misses the RA1
and HY features while RA2 and HBA are satisfied with the presence of a phenyl in R1 and carbonyl
group of the isoxazolidinone ring respectively; 2c satisfies only the RA2 feature, and partially the
HBA feature with the presence of the p-methoxyphenylic oxygen; 2e accomplishes the HY and
partially the RA1 feature; whereas all the 3-D QSAR models are able to frame their level of
activity: for this reason and for the above considerations, the models show a good predictive ability
although different scaffold endowed molecules were used as test set. As examples of the 3-D QSAR
model application the most and least active monocarbamate derivatives (1d and 1a, respectively)
overlapped with the A probe model PLS-coefficients are depicted in Figure 5A, while the most and
the least active dicarbamate molecules (2d and 2f, respectively) are reported in Figure 5B. Despite
the above considerations differences in experimental activities appears to be determined by a better
overlap of the aromatic ring with the p-fluorine on the HY and RA2 regions
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 6. Test Set predictions
P
A
PC
3
SDEPEXT
0.88
C
3
0.88
HD
NA
N
3
3
3
0.81
0.82
0.83
OA
e
3
4
0.84
0.90
d
4
1.51
SDEP values considering the optimal PCs; P: AutoGrid Probe; PC: optimal
number of principal components/latent variables; SDEPEXT: standard deviation
error of prediction (or root mean squared error of prediction, RMSEP) for the
external test set.
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 20 of 37
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
A
B
Figure 5. PLS-coefficients contour maps. A: AutoGrid/R PLS-coefficients contour maps
derived from the A probe analysis (contour levels: 80%; positive: red, negative: blue; 1a:
yellow; 1d: green); B: AutoGrid/R PLS-coefficients contour maps derived from the A probe
analysis (contour levels: 80%; positive: red, negative: blue; 2d: cyan; 2f: orange). HY
(hydrophobic feature), RA (aromatic feature), HBA (hydrogen bond acceptor feature).
ACS Paragon Plus Environment

Page 21 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
Analogously, the MP model predictions were in good agreement with those of the mono probe
models (Table 7). The 3-D QSAR MP plots showing the most active and less monocarbamates and
dicarbamates derivatives (Figure 6) indicated the lack of a simultaneous coverage of the different
regions addressed by the PLS-coefficients. However, as shown in Figure 6 was confirmed the
importance of hydrophobic substituents in the HY and RA2 areas, which should determine the
highest activities of 1d and 2d.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Table 7. MPGRS. Multi Probe model Test Set predictions
P
PCFL:SL
1:3
SDEPEXT
0.89
AutoGrid MP
SDEP values considering the optimal first level and second level PCs. P:AutoGrid Multi-Probe;
PCFL:SL: optimal first level and second level PC; SDEPEXT: standard deviation error of prediction (or
root mean squared error of prediction, RMSEP) for the external test set.
A
B
Figure 6. MPGRS. A: PLS-coefficients contour maps at PC1:3 (contour levels: positive 85%,
red; negative 95%, blue; 1a: yellow;1d: green); B: PLS-coefficients contour maps at PC1:3
(contour levels: positive 85%, red; negative 95%, blue; 2d: cyan; 2f: orange). HY (hydrophobic
feature), RA (aromatic feature), HBA (hydrogen bond acceptor feature).
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 22 of 37
1
2
3
4
5
6
2.2 Chemistry
7
8
9
The synthesis of compounds (1a-e) and (2a-f-) was carried out modifying a literature procedure
described by Stammer and coworkers,20 by treatment of , D-4-amino-3-isoxazolidinone in weakly
alkaline media (1M NaHCO3) with the corresponding chloroformate to obtain both mono- and
dicarbamate derivatives, as illustrated in Scheme 1.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Scheme 1. a) R-O-CO-Cl, 1M NaHCO3, 0°C 3h, 4M HCl; b) R-O-CO-Cl, NaHCO3, 0°C, 15h; c)
0.1 M TEAHFP in CH3CN, 30 mAcm-2, D-4-amino-3-isoxazolidinone 1 eq (15 min), R1-Br; d) 3-
CF3-C6H4COCl, TEA, CHCl3; e) 1f , n-Bu-O-CO-Cl, 1M NaHCO3, 0°C, 24h.
The derivatives 1a-f were prepared by regioselective acylation of the 4-amino group using the
appropriate chloroformates at low temperature for short reaction time (0 °C for 3 h); then pure
solids 1a-f were obtained by acidification with 4M HCl. Dicarbamate derivatives 2a-e were
ACS Paragon Plus Environment

Page 23 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
synthetized dissolving the D-4-amino-3-isoxazolidinone in a basic solution at 0°C and the selected
chloroformates were dropwise added. 2a-e gradually precipitate in 12 h from the acqueous solution.
The compound 2f was obtained by treatment of 1f with n-butylchloroformate in alkaline solution.
N-2-(alkyl)-4-amino-3-oxoisoxazolidinone -4h-i were synthesized by an electrochemical reaction as
previously reported.21 Then crude 4h-i were acylated with 3-trifluoromethyl benzoyl chloride in
chloroform / TEA to give the amide derivatives 3h-i.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
In order to verify the racemization of the α carbon of D-4-amino-3-isoxazolidinone in the reaction
conditions, we have analyzed by chiral HPLC the enantiopurity of (R)-3h and (S)-3h, obtained with
the same synthetic procedure starting from (R)-4-amino-3-isoxazolidinone and (S)-4-amino-3-
isoxazolidinone. Chiralpak Column IC 250 mm x 4.6 mm I.D was used with n-hexane-2-propanol
75/25 (v/v) as eluent at flow rate of 1.0 mL/min at the temperature of 25°C. In both the
chromatogram of (R)-3h and (S)-3h an enantiomeric excess > 99.0 % was observed (Figure S15).
2.3 Biological Activity of Synthesized Compounds
The compounds were assayed for their antimycobacteria activity toward M. tuberculosis H37Rv
(ATCC 27294). The minimal concentration inhibiting visible growth of mycobacteria was
determinated for each compounds.
Concerning the data reported in Table 7 only the acylation of 4-amino group and acylation or
alkylation of N-2 of D-4-amino-3-isoxazolidinone influenced the antitubercular activity leading to
a MIC value of 3.1 μg/mL (as in the case of 1d and 2d). Most of the tested compounds showed the
same activity of the 4-amino-3-isoxazolidinone (32 μg/mL), while 1d and 2d resulted more active
(3.1 μg/mL) and only the compound 3h resulted the less active (64 μg/mL).
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 24 of 37
1
2
3
4
5
6
7
8
9
Table 7. MIC data for D-4-amino-3-isoxazolidinone derivatives
compd
1a
MIC(μg/mL)a
pMICb
3.84
3.87
3.90
4.89
3.97
4.03
4.06
4.10
5.09
4.11
4.02
3.78
4.13
32
32
32
3.1
32
32
32
32
3.1
32
32
64
32
1b
1c
1d
1e
2a
2b
2c
2d
2e
2f
3h
3i
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
aM. tuberculosis H37Rv (ATCC 27294) was used; MIC values represent the minimal
concentrations of compounds completely inhibiting visible growth of mycobacteria.
bpMIC = -Log [MIC(µM) x 10-6]
3. Conclusion
In this paper we present the first application of a quantitative pharmacophoric model able to define
and correlate the needed chemical characteristics with antitubercular activity of a previously
reported class of antimycobacterial agents.8,13-17 Eight 3-D QSAR mono probe models and a multi
probe (MP) model were built showing appreciable statistical coefficients, and allowing an accurate
definition of the structure-activity relationships on the basis of pyrrole derivatives used as training
set. The MP 3-D QSAR model let to define the training set molecular features and their three-
dimensional positioning, configuring itself as a quantitative pharmacophoric model. Furthermore it
was possible to elucidate the effect of conformational differences on the biological activity. As a
further assessment the multi probe information was compared with the original pharmacophoric
model, previously developed by us,14 showing an high degree of correspondence.
Independently, a series of13 isoxazolidinone derivatives 1-3 (Table 1) was synthesized and tested as
new antitubercular compounds. The new compounds showed MIC values in the micromolar range.
In particular among the monocarbamates and dicarbamates, derivatives 1d and 2d showed the
higher biological activities. Although there are limitations due to structural differences between the
ACS Paragon Plus Environment

Page 25 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
6
7
8
9
molecules of the training set and those of derivatives 1-3, the latters were used as an external test set
to evaluate the models’ predictive capabilities. All the 3-D QSAR models showed prediction errors
(Tables 6 and 7), against these structurally unrelated molecules, with an acceptable degree of
approximation.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
The application of the models allowed to clarify the role of halogens and phenyl rings in 1a-e and
2a-f. Considering all these outcomes, the MP 3-D QSAR model could represent a useful tool for the
design of new antitubercular drugs.
4. Experimental Section
4.1 Molecular Modeling and 3-D QSAR
All calculations used a 6 blades (8 Intel-Xeon E5520 2.27 GHz CPU and 24 GB DDR3 RAM each)
cluster (48 CPU total) running Debian GNU/Linux 6.0 64 bit operating system. A series of 71
previously described pyrrole derivatives10, 15, 19 were used to build 8 single probe and a multi probe
3-D QSAR models using the 3-D QSAutogrid/R procedure.11 The obtained models were tested
predicting the activities of the monocarbamates (1a-e), dicarbamates (2a-f) and amides (3h,i) of D-
4-amino-3-isoxazolidinone derivatives.
Training set selection:
10, 15, 19
Starting from a training set composed by 90 pyrrole derivatives,
a selection based on inner
relationship analysis were conducted to improve the robustness and prediction capabilities of the 3-
D QSAR models: this leads to a final training set composed by 71 molecules (Table 2).
Alignment rules:
Training Set: Training set compounds were first submitted to a conformational search following a
computational protocol previously described.10 Next, each compound with its conformational
models was aligned to the pharmacophoric model with the flexible fitting method implemented in
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 26 of 37
1
2
3
4
5
6
Discovery Studio (version 3.0, Accelrys, Inc., San Diego, CA), that allows slight modification of
each conformation to better fit the pharmacophore.
7
8
9
Test Set: The new 13 derivatives were aligned using the Surflex-Sim22 software which has been
chosen since it’s a valuable tool in ligand-based drug discovery, free for academics, and its
alignment process is based on morphological similarities. The query molecules’ poses were
optimized to the compounds used as training set to maximize 3D similarity.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
As shown in Figures S16, S17, Tables S9 and S10, similar results were obtained using the same
alignment software (pharmacophoric alignment) adopted for the training set. The choice of Surflex
alignment was not dictated by the improvement (although negligible respect the pharmacophoric
one) in prediction, but by the fact that, in our view, this is a further confirmation of the robustness
of the models, always taking into account the above limits. Indeed, using two different procedures
of alignment, results are comparable and this should show that the predictive capability of the
models is stable and in the specific case scarcely influenced. Further clarifications on the
differences in prediction (as in the case of 1d and 2d) would be only speculative since the presence
of similar activity values with a limited total pMIC activity range.
Strategies for different alignments were also tried leading to not consistent prediction supporting
that the best alignment/prediction is that above reported.
Molecular Interaction Fields Calculation.
As reported,11 MIFs were generated using the AutoGrid Software (AutoDock Suite,23 based on the
AMBER united-atom Force Field) implemented in the 3-D QSAutogrid/R procedure, considering 8
different probes. Interaction energies between the selected probes and each molecule were
computed using a grid spacing of 1 Å (Tables S8). The xyz coordinates (in angstroms) of the grid
rectangular box used for the computation were Xmin/Xmax = -9.828/12.172, Ymin/Ymax = -
9.021/8.979, Zmin/Zmax = -10.481/9.519.
ACS Paragon Plus Environment

Page 27 of 37
Journal of Chemical Information and Modeling
1
2
3
Statistical Analysis.
4
5
6
7
8
Through the D2M package11 8 3-D QSAR PLS models were built. During the model definition the
assessment of quality and robustness was conducted via two cross-validation (CV package)11
procedures as follows: (1) Leave-One-Out (LOO) and (2) k-Fold (KF, 5-random groups and 100
iterations) methodologies. Initially, the raw models (Tables S2) were optimized through the
Combinatorial Analysis of Pretreatment Parameters (CAPP package)11 setting the pretreatment
intervals as listed in Table 3, using the k-fold cross-validation with 5-random groups and 100
iterations and monitoring the q2 and SDEP values. A total of 726 combinations, for each 3-D QSAR
model, were processed using 5% sPRESS reduction11 to select the optimal pretreatment
combination and derive the pretreated PLS models; this led to an average q2 K5FCV value
increment equal to 14% (Tables S4). Furthermore the scrambling approach, Y-Scrambling,
(package YS)11 was applied to investigate the presence of chance correlations using 100 iterations.
Considering the obtained good overall statistical coefficients together with the absence of chance
correlations (Table 4, TableS3 and Figure S1), no further variable selection steps were performed.
By the application of the MPGRS package,11 a MP 3-D QSAR PLS model was then derived by
selecting the most informative subregions for each of the eight considered probe; the same CV and
scrambling procedures as those of the mono probe models were performed and the optimal MP 3-D
QSAR model was selected according to the q2FL:SLvalues.11 Similar statistical coefficients to those
of the mono probe models were obtained (Table 5, Figure S9) and no further variable selection were
performed; finally the most relevant MIFs subregions were selected applying a q2 threshold value of
0.4.
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
4.2 Chemistry
D-4-amino-3-isoxazolidinone and all chloroformates were purchased from Sigma-Aldrich (Milano,
Italy). All other reagents and solvents were of higher analytical grade. N-benzyloxycarbonyl-D-4-
amino-3-isoxazolidinone (1f) was prepared according to Stammer et al..20 Melting points were
ACS Paragon Plus Environment

Journal of Chemical Information and Modeling
Page 28 of 37
1
2
3
4
5
6
7
8
9
determined on Tottoli apparatus (Buchi) and are uncorrected. Vibrational spectra were recorded on
a Spectrum One ATR Perkin Elmer FT-IR spectrometer. 1H and 13C-NMR spectra were acquired on
a Bruker AVANCE-400 spectrometer at 9.4 Tesla, in DMSO-d6 or CDCl3 at 27°C; chemical shift
values are given in  (ppm) relatively to TMS as internal reference, coupling constants are given in
Hz.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Mass spectra were recorded on: a API-TOF Mariner by Perspective Biosystem (Straford, Texas,
USA), samples were injected by an Harvard pump using a flow rate of 5-10 l/min, infused in the
Electrospray system; a TSQ quadrupole Mass spectrometer by Thermofinnigan (S. Jose, California,
USA) operating in CH4 / CI conditions, samples were introduced in the CI source by a direct
insertion probe. Elemental analyses were obtained by a PE 2400 (Perkin-Elmer) analyzer.
General procedure for synthesis of monocarbamates 1a-e
Compounds 1a-e were prepared by a modified procedure described by Stammer et al.,20: briefly, to
0.5 mmol of D-4-amino-3-isoxazolidinone, dissolved in 1.2 mL of aqueous 1M NaHCO3, cooled in
an ice bath, 1.0 mmol of the corresponding chloroformate was added and the solution stirred for 3
hours at 0°C. Little amounts of precipitate that could be formed were filtered off and the solution,
kept to 0°C, was acidified to pH 4 with 4M aqueous HCl. After 30 min, the white fine precipitate
was collected by centrifugation and washed with cold water. The collected carbamates 1a-e were
crystallized from water.
Synthesis of (R)-phenyl-(3-oxoisoxazolidin-4-yl)carbamate (1a)
1a Was obtained as white crystalline solid in 75% yield, mp 148- 9°C. 1H NMR (DMSO-d6): 11.52
(1H, bs, D2O exchange); 8.37 ( 1H, s, D2O exchange); 7.38 ( t, 2H, J = 8.56 Hz); 7.21 ( t,1H, J =
7.58 Hz); 7.12 ( d, 2H, J = 8.56 Hz); 4.65-4.60, ( m, 1H); 4.56 ( t, 1H, J = 9.78 Hz); 4.03 (t,1H, J =
8.32 Hz). 13C NMR (DMSO-d6): 170.5; 154.9; 151.4; 129.9; 125.9; 122.3; 72.2; 53.5.FT-IR (cm-1):
3339, 1709, 1655. MS/ESI: (M+H)+ 223.0730 (m/z).
ACS Paragon Plus Environment

Page 29 of 37
Journal of Chemical Information and Modeling
1
2
3
4
5
Synthesis of (R)-4-methylphenyl-(3-oxoisoxazolidin-4-yl)carbamate (1b)
6
7
8
9
1b Was obtained as white crystalline solid in 75% yield, mp 195-6°C. 1H NMR (DMSO-d6): 11.54
(bs, 1H, D2O exchange); 8.30 (s, 1H, D2O exchange); 7.16 (d, 2H, J = 8.56 Hz); 6.99 (d, 2H, J =
8.56 Hz); 4.66-4.59 (m, 1H); 4.55 (t, 1H, J = 10.35 Hz); 4.02 (t, 1H, J= 8.32 Hz); 2.28 (s, 3H). 13C
NMR (DMSO-d6): 169.9; 154.4; 148.5; 134.2; 129.6; 121.3; 71.6; 52.8; 20.3. FT-IR (cm-1): 3305,
1709, 1654. MS/ESI: (M+H)+ 237.0874 (m/z).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Synthesis of (R)-4-methoxyphenyl-(3-oxoisoxazolidin-4-yl)carbamate (1c)
1c Was obtained as white crystalline solid in 70% yield, mp 155-6°C. 1H NMR (DMSO-d6): 11.45
(bs, 1H, D2O exchange); 8.27 (s, 1H, D2O exchange); 7.02 (d, 2H, J = 9.06 Hz); 6.91 (d, 2H, J =
13
9.06 Hz); 4.64-4.59 (m, 1H); 4.54 (t, 1H, J = 9.29 Hz); 4.02 (t, 1H, J = 9.29 Hz); 3.73 (s, 3H). C
NMR (DMSO-d6): 170.0; 156.5; 154.6; 144.2; 122.5; 114.2; 71.6; 55.4; 52.8. FT-IR (cm-1): 3314,
1709, 1655. MS/ESI: (M+H)+ 253.0790 (m/z).
Synthesis of (R)-4-fluorophenyl-(3-oxoisoxazolidin-4-yl)carbamate (1d)
1d Was obtained as white crystalline solid in 70% yield, mp 175-6°C. 1H NMR (DMSO-d6): 11.54
(bs,1H, D2O exchange); 8.39 (s,1H, D2O exchange); 7.16 (d, 2H, J = 8.56 Hz); 6.98 (d,2H, J = 8.56
Hz); 4.68-4.63 (m,1H ); 4.55 (t, 1H, J = 9.56 Hz); 4.03 (t,1H, J = 8.80 Hz). NMR (DMSO-d6):
155.9 (d, J= 237.8 Hz); 154.1; 147.3; 123.9 (d, J= 7.32 Hz); 116.5; 116.0 (d, J= 22.7 Hz); 72.0;
53.4. FT-IR (cm-1): 3320, 1712, 1699. MS/CI: (M+H)+ 241 (m/z).
Synthesis of (R)-4-bromophenyl-(3-oxoisoxazolidin-4-yl)carbamate (1e)
1e Was obtained as white crystalline solid in 75% yield, mp 180-1°C. 1H NMR (DMSO-d6): 11.55
(bs,1H, D2O exchange); 8.42 (s, 1H, D2O exchange); 7.22 (d, 2H, J = 8.56 Hz); 7.16 (d, 2H, J =
8.56 Hz); 4.69- 4.62 (m, 1H); 4.56 (t,1H, J = 9.88 Hz); 4.04 (t, 1H, J = 8.80 Hz). 13C NMR (DMSO-
ACS Paragon Plus Environment