Synthesis, molecular modeling studies and evaluation of antifungal activity of a novel series of thiazole derivatives

Article abstract of DOI:10.1016/j.ejmech.2018.03.083

In the search for new antifungal agents, a novel series of fifteen hydrazine-thiazole derivatives was synthesized and assayed in vitro against six clinically important Candida and Cryptococcus species and Paracoccidioides brasiliensis. Eight compounds showed promising antifungal activity with minimum inhibitory concentration (MIC) values ranging from 0.45 to 31.2 μM, some of them being equally or more active than the drug fluconazole and amphotericin B. Active compounds were additionally tested for toxicity against human embryonic kidney (HEK-293) cells and none of them exhibited significant cytotoxicity, indicating high selectivity. Molecular modeling studies results corroborated experimental SAR results, suggesting their use in the design of new antifungal agents.

Full text of DOI:10.1016/j.ejmech.2018.03.083

Accepted Manuscript
Synthesis, molecular modeling studies and evaluation of antifungal activity of a novel
series of thiazole derivatives
Cleudiomar Inácio Lino, Igor Gonçalves de Souza, Beatriz Martins Borelli, Thelma
Tirone Silvério Matos, Iasmin Natália Santos Teixeira, Jonas Pereira Ramos, Elaine
Maria de Souza Fagundes, Philipe de Oliveira Fernandes, Vinícius Gonçalves
Maltarollo, Susana Johann, Renata Barbosa de Oliveira
PII:
S0223-5234(18)30326-X
10.1016/j.ejmech.2018.03.083
EJMECH 10348
DOI:
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 12 October 2017
Revised Date: 2 March 2018
Accepted Date: 30 March 2018
Please cite this article as: Cleudiomar.Iná. Lino, I. Gonçalves de Souza, B.M. Borelli, T.T. Silvério Matos,
Iasmin.Natá. Santos Teixeira, J.P. Ramos, E. Maria de Souza Fagundes, P. de Oliveira Fernandes,
Viní.Gonç. Maltarollo, S. Johann, R.B. de Oliveira, Synthesis, molecular modeling studies and evaluation
of antifungal activity of a novel series of thiazole derivatives, European Journal of Medicinal Chemistry
(2018), doi: 10.1016/j.ejmech.2018.03.083.
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

1
ACCEPTED MANUSCRIPT
Synthesis, Molecular Modeling Studies and Evaluation of Antifungal Activity of a Novel
Series of Thiazole Derivatives
Cleudiomar Inácio Linoa, Igor Gonçalves de Souzaa, Beatriz Martins Borellib, Thelma Tirone
Silvério Matosb, Iasmin Natália Santos Teixeiraa, Jonas Pereira Ramosc, Elaine Maria de
Souza Fagundesc, Philipe de Oliveira Fernandesd, Vinícius Gonçalves Maltarolloa, Susana
Johannb, Renata Barbosa de Oliveiraa*.
aDepartamento de Produtos Farmacêuticos, Faculdade de Farmácia, Universidade Federal de
Minas Gerais, Belo Horizonte, MG, Brazil.
bDepartamento de Microbiologia, Instituto de Ciências Biológicas, Universidade Federal de
Minas Gerais, Belo Horizonte, MG, Brazil.
cDepartamento de Fisiologia e Biofísica, Instituto de Ciências Biológicas, Universidade
Federal de Minas Gerais, Belo Horizonte, MG, Brazil.
dDepartamento de Química, Instituto de Ciências Exatas, Universidade Federal de Minas
Gerais, Belo Horizonte, MG, Brazil.
*Corresponding author: Tel.: +55-31-34096395; fax: +55-31-34096935. E-mail address:
renatabo@farmacia.ufmg.br

2
ACCEPTED MANUSCRIPT
Abstract
In the search for new antifungal agents, a novel series of fifteen hydrazine-thiazole derivatives
was synthesized and assayed in vitro against six clinically important Candida and
Cryptococcus species and Paracoccidioides brasiliensis. Eight compounds showed promising
antifungal activity with minimum inhibitory concentration (MIC) values ranging from 0.45 to
31.2 µM, some of them being equally or more active than the drug fluconazole and
amphotericin B. Active compounds were additionally tested for toxicity against human
embryonic kidney (HEK-293) cells and none of them exhibited significant cytotoxicity,
indicating high selectivity. Molecular modeling studies results corroborated experimental
SAR results, suggesting their use in the design of new antifungal agents.
Keywords: antifungal agents; hydrazine-thiazoles; QSAR

3
ACCEPTED MANUSCRIPT
1. Introduction
Over the last decades, there has been a significant increase in the incidence of fungal
infections in humans, mainly affecting immunocompromised patients [1,2]. Among the fungi
of clinical importance, pathogenic yeasts of the genus Candida and Cryptococcus are the most
relevant [3], since candidemia is the third or fourth most common cause of healthcare-
associated bloodstream infections in US hospitals [4]. Also in Brazil, mortality rates of
candidemia can exceed more than half of the clinical cases [5,6]. The main pathogenic species
of the genus Candida are C. albicans, C. parapsilosis, C. tropicalis, C. glabrata, C. krusei, C.
guilliermondii and C. lusitaniae. Among those, C. albicans is the most frequently isolated
agent of candidiasis, but non-albicans species of Candida have also increased their prevalence
and gained clinical significance worldwide in the last years [7,8].
Cryptococcosis is an infectious disease with worldwide distribution and a wide array of
clinical presentations caused by pathogenic encapsulated yeasts in the genus Cryptococcus,
mainly C. neoformans and C. gattii [9]. Cryptococcal meningitis is one of the most important
HIV related opportunistic infections and it is associated with significant mortality,
approximately 181,100 deaths per year [10].
Paracoccidioidomycosis (PCM) is a disease mostly limited to Latin America and is caused by
one of the two known species of the genus Paracoccidioides, P. brasiliensis and P. lutzii.
PCM is the highest cause of mortality among systemic mycoses in Brazil, causing 1.65 deaths
per 106 inhabitants [11-13].
The treatment of fungal infections is very limited, when compared to bacterial infections, with
only a limited number of available drugs. This is attributable to the nature of the fungal's cell
that, as eukaryotic organisms, contain few drug targets not shared with human hosts [14].
Although much progress has been achieved in the development of therapeutic alternatives for
treatment of these infections, the search for new antifungal compounds is still mandatory
because of the serious side effects of available drugs and emergence of resistance [15].
Thiazole heterocycles constitute an interesting class of molecules, which exhibit a broad
spectrum of biological activity, including antifungal properties [16-19]. Encouraged by our
previous study that describes the activity of hydrazine-thiazole derivatives against C. gattii
and C. neoformans [20], we decided to synthesize novel compounds of this class in order to
investigate their activity against different species of fungi.
Additionally, 2D- and 3D-QSAR studies, as well as 3D similarity models, were constructed
for prediction of biological activities and provide a basis for the design of novel compounds
having improved potency.

4
ACCEPTED MANUSCRIPT
2. Materials and Methods
2.1 Chemistry - All melting points were determined on a Microquímica MQAPF 301
apparatus. The IR spectra were recorded using a PerkinElmer Spectrum One infrared
spectrometer and absorptions are herein reported as wave numbers (cm-1). The NMR spectra
were recorded on a Bruker AVANCE DRX200 instrument, using tetramethylsilane (TMS) as
the internal standard. All reagents of analytical grade were obtained from commercial
suppliers and used without further purification. The synthesis and the characterization of the
ketones 16a [21] and 17a [22] and thiosemicarbazones 16b-22b were previously reported in
the literature [23-29], as well as the synthesis of the hydrazine-thiazoles 10 [26] and 11 [30].
2.1.1 General procedure for synthesis of hydrazine-thiazoles 1-15 [31]
To a solution of 1 equivalent of thiosemicarbazones 16b-22b in isopropyl alcohol was added
1 equivalent of non-substituted or para-substituted 2-bromoacetophenone. The reaction
mixture was kept under reflux and magnetic stirring for approximately 90 minutes. After the
completion of the reaction (monitored by TLC), the mixture was then cooled to room
temperature for precipitation. The precipitate was filtered and washed with saturated solution
of NaHCO3 followed by cold distilled water.
2.1.1.1 Synthesis of 2-[(2-(hexan-3-ylidene)hydrazo]-4-phenylthiazole (1)
Using the general procedure described above (section 2.1.1), 1 was obtained as a yellow solid
in 94 % yield. Mp: 105.1-106.5 °C; IR (cm-1): 2960, 2933, 2869, 1616, 1493, 1450; 1H NMR
(200 MHz, CDCl3), δ/ppm: 13.8 (1H, s), 12.8 (1H, s), 7.7 (2H, m), 7.5 (3H, m), 6.7 (1H, s),
2.6-2.3 (4H, m), 1.8-1.6 (2H, m), 1.3-1.0 (6H, m); 13C NMR (50 MHz, CDCl3), δ/ppm: 170.1,
162.3, 145.0, 130.7, 129.2, 125.8, 102.0, 32.8, 30.0, 19.2, 14.3, 10.5; HRMS (m/z) 374.1372
[M+H]+, calcd274.1378 C15H20N3S+.
2.1.1.2 Synthesis of 2-[(2-(hexan-3-ylidene)hydrazo]-4-(4-methoxyphenyl)-thiazole (2)
Using the general procedure described above (section 2.1.1), 2 was obtained as a pale solid in
97% yield. Mp: 106.8-107.4 °C; IR (cm-1): 3504, 2960, 2934, 1607, 1590, 1508, 1455, 1256,
1
1188; H NMR (200 MHz, CDCl3), δ/ppm: 13.7 (1H, s), 12.4(1H, s), 7.6(2H, d), 6.9(2H, d),
13
6.5(1H, s), 3.8(3H, s), 2.6-2.3 (4H, m), 1.7-1.6(2H, m), 1.3-1.1(6H, m); C NMR(50 MHz,
CDCl3), δ/ppm: 170.0, 165.2, 161.0, 141.8, 127.3, 121.3, 114.9, 99.3, 55.5, 38.0, 30.0, 33.3,
24.6, 19.4, 14.2, 13.7, 10.4, 10.1; HRMS (m/z) 304.1485 [M+H]+, calcd304.1487
C16H22N3OS+.

5
ACCEPTED MANUSCRIPT
2.1.1.3 Synthesis of 2-[(2-(hexan-3-ylidene)hydrazo]-4-(4-chlorophenyl)-thiazole (3)
Using the general procedure described above (section 2.1.1), 3 was obtained as a pale solid in
99 % yield. Mp: 148.2-149.1 °C; IR (cm-1): 3156, 3108, 2961, 2931, 2871, 1558, 1477, 1457,
731; 1H NMR (200 MHz, DMSO-d6), δ/ppm: 13.8 (1H, s), 12.4 (1H, s), 7.7 (2H, d), 7.4 (2H,
d), 6.8(1H, s), 2.6-2.3(4H, m), 1.7-1.6 (2H, m), 1.3-1.0 (6H, m); 13C NMR (50 MHz, DMSO-
d6), δ/ppm: 170.1, 167.3, 139.3, 136.5, 129.9, 127.0, 125.9, 101.8, 38.1, 30.1, 33.6, 25.0,
19.5, 19.4, 14.3, 13.8, 10.3; HRMS (m/z) 308.0983 [M+H]+, calcd 308.0988 C15H18ClN3S+.
2.1.1.4 Synthesis of (E)-2-[2-(2,2-dimethylpropylidene)hydrazo]-4-phenylthiazole (4)
Using the general procedure described above (section 2.1.1), 4 was obtained as a violet solid
in 90% yield. Mp: 101.2-102.1 °C; IR (cm-1): 3062, 2959, 2901, 2865, 1626, 1603, 1584,
1576, 1483, 1443; 1H NMR (200 MHz, DMSO-d6), δ/ppm: 7.8-7.7 (3H, m), 7.4-7.3 (4H, m),
13
7.2 (1H, s), 1.1 (9H, s); C NRM (50 MHz, DMSO-d6), δ/ppm: 169.1, 154.9, 148.1, 133.5,
128.7, 127.9, 125.7, 103.5, 34.6, 27.3; HRMS (m/z) 260.1216 [M+H]+, calcd 260.1221
C14H18N3S+.
2.1.1.5 Synthesis of (E)-2-[2-(2,2-dimethylpropylidene)hydrazo]-4-(4-methoxyphenyl)-
thiazole (5)
Using the general procedure described above (section 2.1.1), 5 was obtained as a pale solid in
94% yield. Mp: 104.5-105.8 °C; IR (cm-1): 3061, 2960, 2836, 1624, 1574, 1510, 1493, 1462,
1
1248; H NMR (200 MHz, DMSO-d6), δ/ppm: 7.7 (2H, d); 7.5 (1H, s); 7.1 (1H, s); 6.9 (2H,
d); 3.8 (3H, s); 1.1 (9H, s);13C NMR (50 MHz, DMSO-d6), δ/ppm: 169.1, 159.3, 155.9,
146.7, 127.2, 125.5, 114.1, 101.6, 55.2, 34.7, 27.2; HRMS (m/z) 290.1322 [M+H]+, calcd
290.1327 C15H20N3OS+.
2.1.1.6 Synthesis of (E)-2-[2-(2,2-dimethylpropylidene)hydrazo]-4-(4-chlorophenyl)-
thiazole (6)
Using the general procedure described above (section 2.1.1), 6 was obtained as a pale solid in
87 % yield. Mp: 148.5-149.6 °C; IR (cm-1): 3162, 3116, 3078, 2963, 2863, 1568, 1479, 727;
1H NMR (200 MHz, DMSO-d6), δ/ppm: 11.6 (1H, s), 7.8 (2H, d), 7.4 (2H, d), 7.3(1H, s), 7.2
(1H, s), 1.1 (9H, s); 13C NRM (50 MHz, DMSO-d6), δ/ppm: 169.0, 152.9, 149.0, 133.6,
131.8, 128.5, 127.2, 103.8, 34.4, 27.3; HRMS (m/z) 294.0826 [M+H]+, calcd 294.0832
C14H17ClN3S+.
2.1.1.7 Synthesis of (E)-3-buten-2-one-4-(4-hydroxy-3-methoxyphenyl)-2-(4-phenyl-2-
thiazolyl)hydrazone (7)

6
ACCEPTED MANUSCRIPT
Using the general procedure described above (section 2.1.1), 7 was obtained as an orange
solid in 76 % yield. Mp: 155.5-156.5 °C; IR (cm-1): 3485, 3069, 2965, 1615, 1574, 1516,
1489, 1282, 1266; 1H NMR (200 MHz, DMSO-d6), δ/ppm: 11.1 (1H, s), 9.2 (1H, s), 7.8 (2H,
d), 7.4-7.1 (5H, m), 7.0-6.8 (3H, m), 3.8 (3H, s), 2.1 (3H, s); 13C NRM (50 MHz, DMSO-d6),
δ/ppm: 169.2, 150.4, 148.7, 147.9, 147.2, 134.8, 132.3, 128.6, 128.0, 127.4, 125.5, 120.7,
115.6, 110.0, 103.8, 55.6, 12.3; HRMS (m/z) 366.1269 [M+H]+, calcd 366.1271
C20H20N3O2S+.
2.1.1.8
Synthesis
of
(E)-3-buten-2-one-4-(4-hydroxy-3-methoxyphenyl)-2-(4-(4-
methoxyphenyl)-2-thiazolyl)hydrazone (8)
Using the general procedure described above (section 2.1.1), 8 was obtained as an orange
solid in 82 % yield. Mp: 188-189.6 °C; IR (cm-1): 3202, 3113, 2927, 2837, 1609, 1510, 1457,
1
1256; H NMR (200 MHz, DMSO-d6), δ/ppm: 11.1 (1H, s), 9.2 (1H, s), 7.8 (2H, d), 7.4-7.3
(2H, m), 7.2-6.9 (3H, m), 6.8-6.6 (3H, m), 3.8 (3H, s), 3.7 (3H, s), 2.1 (3H, s); 13C NRM (50
MHz, DMSO-d6), δ/ppm: 169.0, 158.7, 150.2, 148.6, 147.8, 147.1, 132.2, 128.0, 127.6,
126.8, 125.7, 120.6, 115.6, 113.9, 110.0, 101.6, 55.6, 55.1, 12.2; HRMS (m/z) 396.1378
[M+H]+, calcd 396.1376 C21H22N3O3S+.
2.1.1.9
Synthesis
of
(E)-3-buten-2-one-4-(4-hydroxy-3-methoxyphenyl)-2-(4-(4-
chlorophenyl)-2-thiazolyl)hydrazone (9)
Using the general procedure described above (section 2.1.1), 9 was obtained as an orange
solid in 88 % yield. Mp: 166.2-166.8 °C; IR (cm-1): 3487, 2964, 1616, 1591, 1513, 1489,
1
1280, 1257, 755; H NMR (200 MHz, DMSO-d6), δ/ppm: 11.1 (1H, s), 9.2 (1H, s), 7.9 (2H,
d), 7.4 (1H, d), 7.3 (1H, d), 7.2 (1H, s), 6.98 (1H, d), 6.91 (1H, d), 6.7-6.6 (2H, m), 3.8 (3H,
s), 2.1 (3H, s); 13C NRM (50 MHz, DMSO-d6), δ/ppm: 169.7, 149.6, 149.2, 148.3, 147.6,
134.1, 132.8, 132.2, 128.9, 128.4, 127.6, 126.0, 121.1, 115.9, 110.4, 105.0, 50.0, 12.6; HRMS
(m/z) 400.0881 [M+H]+, calcd 400.0881 C20H19ClN3O2S+.
2.1.1.10 Synthesis of (E)-2-[(2-(heptan-2-ylidene)hydrazo]-4-(4-cyanophenyl)-thiazole
(12)
Using the general procedure described above (section 2.1.1), 12 was obtained as a pale solid
1
in 46 % yield. Mp: 188.2-192.7 °C; H NMR (200 MHz, CDCl3), δ/ppm: 12.2 (1H, s), 7.8
(2H, d, J = 8.4 Hz), 7.7 (2H, d, J = 8.4 Hz), 7.0 (1H, s), 2.4 (2H, m), 2.2 (3H, m), 1.6 (2H, m),
1.3 (4H, m), 0.9 (3H, m); 13C NRM (50 MHz, CDCl3), δ/ppm: 169.8, 163.5, 138.3, 133.2,

7
ACCEPTED MANUSCRIPT
131.1, 126.1, 113.7, 104.3, 38.2, 31.1, 25.2, 22.2, 18.1, 13.8; HRMS (m/z) 313.1491 [M+H]+,
calcd 313.1481 C17H21N4S+.
2.1.1.11
Synthesis
of
(E)-4-(4-chlorophenyl)-2-[2-(1-(4-
aminophenyl)ethylidene)hydrazinyl]-1,3- thiazole (13)
Using the general procedure described above (section 2.1.1), 13 was obtained as a brown solid
in 85 % yield. Mp: 192.8-194.2 °C; 1H NMR (200 MHz, Acetone-d6), δ/ppm: 7.9 (2H, d), 7.6
(2H, d), 7.4 (2H, d), 7.2 (1H, s), 6.7 (2H, d), 2.4 (3H, s); HRMS (m/z) 343.0787 [M+H]+,
calcd 343.0779 C17H16ClN4S+.
2.1.1.12
Synthesis
of
(E)-2-[(2-benzylidenecyclohexylidene)hydrazinyl]-4-
(cianophenyl)thiazole (14)
Using the general procedure described above (section 2.1.1), 14 was obtained as a red solid in
1
72 % yield. Mp: 169.7-171.2 °C; H NMR (200 MHz, DMSO-d6), δ/ppm: 8.0 (2H, d), 7.8
(2H, d), 7.6 (1H, s), 7.4-7.2 (6H, m), 7.0 (1H, s), 2.6-2.5 (4H, m), 1.7-1.5 (4H, m); HRMS
(m/z) 385.1490 [M+H]+, calcd 385.1481 C23H21N4S+.
2.1.1.13
Synthesis
of
(E)-2-propenal-3-phenyl-2-[4-(4-chlorophenyl)2-
thiazolyl]hydrazone (15)
Using the general procedure described above (section 2.1.1), 15 was obtained as a pink solid
in 91 % yield. Mp: 194.8-196.6 °C; 1H NMR (200 MHz, DMSO-d6), δ/ppm: 8.1-8.0 (1H, dd),
7.9-7.8 (2H, d), 7.6 (1H, s), 7.7-7.5 (2H, m), 7.5-7.1 (7H, m), 7.1-6.9 (2H, m); HRMS (m/z)
340.0678 [M+H]+, calcd 340.0670 C18H15ClN3S+.
2.2. Biological assay
2.2.1 Fungal strains and inoculum quantification
For the antifungal evaluation, the following strains from the American Type Culture
Collection (ATCC, Rochville, MD, USA) were used: Cryptococcus gattii (ATCC32608), C.
neoformans (ATCC24067), Candida albicans ATCC 18804, C. krusei (ATCC 20298), C.
parapsilosis (ATCC 22019), C. tropicalis (ATCC 750) and Paracoccidioides brasiliensis
(Pb18). All fungi strains were obtained from the fungi collection of the Faculty of Medicine
of the Universidade de São Paulo (São Paulo, SP, Brazil). The fungi strains were maintained
on Sabouraud Dextrose Agar (SDA, Oxoid, Basingstoke, UK) for most species, and YPD
(Yeast, Peptone and Dextrose) for P. brasiliensis. All yeast strains were stored frozen at -
80ºC.

8
ACCEPTED MANUSCRIPT
Final inocula for Candida and Cryptococcus species were of 1.5 x 103 CFU/mL and were
prepared using the spectrophotometric dilution method. Broth microdilution testing was
performed in accordance with the guidelines in the CLSI M27-A3 document [32], with
modifications proposed by Johann et al. [33]. To inoculate the cultures of P. brasiliensis yeast
cells were aseptically collected with a bacteriological loop and suspended in 5 mL of sterile
saline solution 0.9%. The suspension was homogenized by vortexing for approximately 10
seconds. Suspensions containing larger aggregates of cells were rested before decantation, and
then, only the supernatant was collected. The transmittance was measured at a wavelength of
530 nm and then adjusted to 70%, which corresponded to 1-5 x 106 cells/mL [34]. The
resulting suspensions were diluted in RPMI medium supplemented with L-glutamine and
buffered to pH 7.0 with 0.165 M morpholine propanesulfonic acid (MOPS) (Sigma, St Louis,
USA), broth (1:10) to obtain a final inoculum with 1-5 x 105yeasts/Ml [35].
2.2.2 Determination of minimum inhibitory concentration (MIC)
Broth microdilution testing was performed in accordance with the guidelines in the CLSI
document M27-A3 [32]. Susceptibility towards our test compounds was determined by the
microbroth dilution method, which was performed in sterile flat-bottom 96-well microplates
(Difco Laboratories, Detroit, MI, USA). The compounds were dissolved in DMSO (Vetec,
RJ, Brazil) and diluted in synthetic RPMI medium to obtain a final concentration of 250 µM.
Compounds with activity at this concentration underwent a new test in to determine their
potency (MIC values). The antifungal candidates were tested at concentrations of 0.45-250
µM, with pure RPMI media as growth and sterile control. Fluconazole (0.125 a 64 µg/mL),
amphotericin B and itraconazole (0.03 a 16 µg/mL) were included as the positive antifungal
controls. After inoculation of fungal strains, the plates were incubated at 35 °C during 48 h for
the Candida species, 72 h for Cryptococcus spp. and 7 days for P. brasiliensis. The endpoints
were determined visually by comparison with the endpoints of the drug-free growth-control
wells. All tests were performed in triplicate. The value of the minimum inhibitory
concentration (MIC) was defined as the lowest compound concentration (µM) at which the
well was optically clear.
2.2.3 Cytotoxicity assay in human cells
Human non-tumor embryonic kidney HEK 293 cells were kindly provided by Dr. Marcel Leist,
University of Konstanz / Germany. These cells were cultured in high glucose DMEM medium
(Sigma Aldrich, USA) supplemented with both 10% fetal bovine serum (GIBCO BRL, Grand
Island, NY) and 1% antibiotic solution (100 IU/mL penicillin and 100 µg/mL streptomycin

9
ACCEPTED MANUSCRIPT
(GIBCO BRL, Grand Island, NY). Cells were maintained in the logarithmic growth phase in
incubator with 5% CO2 at 37°C.
Cytotoxicity in HEK 293 cells was assessed by the MTT assay as described by Mosmann [36].
Briefly, HEK cells were plated at the density of 1 x 104 cells per well in 96-well plate and
incubated overnight. After incubation, the cells were treated for 72 hours with thiazole
derivatives, amphotericin B, fluconazole and itraconazole using seven serial dilutions between
250-0.016 µM. Cell viability was evaluated by the rate of reduction of 3-(4,5-dimethylthiazol-
2-yl) -2,5-diphenyltetrazolium bromide (MTT) to formazan crystals quantified by absorbance
at 570 nM in a spectrophotometer (VersaMax). The IC50 was calculated by non-linear
regression using GraphPad Prism® Version 5.01 software (GraphPad Software Inc., La Jolla,
CA, USA). The results were expressed as the percentage of viability in relation to the negative
control (DMSO, 0.5%), which was calculated as follows: Percentage of cell viability (%) =
[(treated mean OD / negative control) x100] and obtained with two independent experiments
performed in triplicate.
2.3 Molecular modeling studies
2.3.1 Dataset compounds
Twenty-two compounds with C. neoformans minimum inhibitory concentration (MIC) were
employed in the 2D- and 3D-QSAR studies. The dataset comprises thiazole derivatives
retrieved from the literature [20], and those synthesized in this work. All dataset compounds
were tested for antifungal activity using the same experimental conditions, allowing the
construction of QSAR models due to the equivalence of biological data. MIC values were
converted into pMIC (-LogMIC) for the inference of the statistical models.
Initially, 2D structures of the dataset compounds were constructed using Discovery Studio
Visualizer [37] and the 3D lowest energy conformation was generated by OMEGA 2.5.1.4
[38,39]. After, all compounds were aligned in common scaffold (hydrazine-thiazole and
benzene ring) using lowest energy conformation of the (E)-3-[2-(4-(4-methoxyphenyl)thiazol-
2-yl)hydrazono)]butan-1-ol [20]. Dataset was separated into two subsets: a training set (80%
of a total number of compounds) which was employed in the model inference and used for
internal validation) and a test set (20% of the complete set) which was used for predictions
and external validation. Three features were used for generating individual hierarchical cluster
analysis (HCA): antifungal space (values of pMIC for C. neoformans inhibition), drug-like
space (represented by molecular weight, ClogP, number of hydrogen bond donors and
acceptors, fraction of sp3 carbons, and number of rotatable bonds), and 2D molecular
similarity (represented by MACC fingerprints). All descriptors from drug-like and 2D
similarity spaces were calculated with PaDEL-Descriptor [40]. HCA was performed

10
ACCEPTED MANUSCRIPT
considering the incremental-linkage clustering method and the normalized Euclidian distance
calculated between each pair. All HCA calculations were performed by using Chemoface
software [41]. After clustering, 20% of compounds were randomly selected for each cluster of
all three HCA, in order to ensure that training and test set compounds would remain
representative of the total dataset considering its biological activity, physicochemical
properties, and molecular structure.
2.3.2 Construction and validation of HQSAR model
After construction of the 3D molecular structure of dataset compounds and its rational
division in training/test sets, hologram QSAR (HQSAR) models were generated in order to
predict C. neoformans inhibitory activities. All HQSAR models were constructed using
training set compounds and Sybyl X 2.1 package [42]. First, 32 HQSAR models were
constructed using different combinations of fragment distinction parameters, which
constitutes the parameters employed to generate the molecular hologram: atoms [A], bonds
[B], connections [C], hydrogen atoms [H] and hydrogen bonds acceptor/donor [DA]. This
initial screening employed fragment size fixed to contain 4 to 7 atoms. The three most robust
HQSAR models (highest q2 values among their pairs) were selected to test the influence of
fragment size in statistical parameters. At that point, twenty-seven new HQSAR models were
constructed by fixing the fragment distinction and then varying the fragment size as 1-4
atoms, 2-5 atoms, 3-6 atoms, 5-8 atoms, 6-9 atoms, 7-10 atoms, 8-11 atoms, 9-12 atoms and
10-13 atoms. In both stages of HQSAR model construction (fragment distinction and
fragment size variation), models with different hologram lengths were generated (from 53 to
401 bins according to HQSAR default parameters).
2.3.3 Construction of 3D-QSAR models
Comparative Molecular Fields Analysis (CoMFA) and Comparative Molecular Similarity
Index Analysis (CoMSIA) were employed as chosen 3D-QSARs approaches. Both studies
were carried out with Sybyl X 2.1 package. Before 3D- QSAR studies, atomic charges were
calculated with the PM3 method using MOPAC software implemented in Sybyl X 2.1
package. Then, CoMFA and CoMSIA models were first constructed using default parameters,
which are energy cutoff set as 30 kcal/mol, a sp3 carbon atom positively charged used as a
probe and grid generated with 2 Å of distance between each grid point. In parallel, CoMSIA
models were constructed from different combinations of similarity indexes (electrostatic,
steric, hydrophobic, H-bond acceptor and H-bond donor fields). After construction of the first
CoMFA and CoMSIA models, the region focusing technique was applied to select most
relevant variables and to obtain more robust models: the distance between grid points was

11
ACCEPTED MANUSCRIPT
varied from 0.5 to 3 Å and a standard deviation factor obtained from first models was applied
to correct calculated fields.
2.3.4 Statistical Validations of 2D- and 3D-QSAR models
All HQSAR, CoMFA and CoMSIA models were evaluated according to i) internal validation
coefficient calculated with leave-one-out method (q2), ii) regression coefficient (r2) and iii) its
respective errors (standard error of validation [SEV] and standard error of estimation [SEE]).
The three most robust models were selected for external validations with test set compounds
and the following metrics were analyzed: i) external validation coefficient (r2test), ii) r2 , a
m
coefficient for comparison between adjusted external validation coefficient and external
validation coefficient forced to pass through the origin (r2 ), r2 ’, the same coefficient
0
m
calculated with inverted axis and it’s iii) average (avgr2 ) and iv) difference ( r2 ) [43], v)
∆
m
m
Q2
and vi) Q2
metrics, and vii) Concordance Correlation Coefficient (CCC) was
(f3)
(f2)
employed for check the correlation between the precision (fitting of observed data with fitting
line) and accuracy (how deviated the regression line is the slope 1 when intercepting origin
point) [44]. Error metrics as RMSEP and MAE were also evaluated as external validations.
The most predictive HQSAR, CoMFA and CoMSIA models were submitted to leave-N-out
internal validation, where N size ranged from 5 to 20 groups of compounds. The compounds
of each N sized group were randomly selected by the validation algorithm and to disregard the
bias, the validation was performed in triplicate. Twenty runs of progressive-scrambling
validations were also performed to the most predictive models to verify chance correlation of
the constructed model. Progressive-scrambling is similar to the Y-scrambling procedure but
performs a scrambling of pMIC values in different activities ranges. In this work, we selected
2 distinct ranges of pMIC (complete dataset and a dataset divided into two groups according
to average pMIC values [called 50/50 subset]) to generate 10 scrambled models for each
subset division. Then, models generated with scrambled activities of 50/50 subset would
produce better validation metrics than full scrambled models [45].
2.3.5 Construction and validation of similarity models
In addition to QSAR models, we constructed 3D similarity models for classification of the
studied dataset into active/inactive classes. Five compounds with highest MIC values were
labeled as active, while the other compounds were considered inactive. Furthermore, 50
decoys (putative inactive compounds) were generated for each active compound using DUD-
E web-server [46]. Then, vROCS software [47,48] was employed to construct and validate
chemical similarity models. Initially, ROCS models were constructed considering all

12
ACCEPTED MANUSCRIPT
chemical features of template compounds, which was followed by a screening of feature
exclusion and weight applying was performed aiming to verify the predictability of models.
After construction of models, 30 conformers of dataset compounds (active, inactive and
decoys) were generated with OMEGA software and models were applied to rank substances
according to similarities, evaluated by shape (volume) and color (chemical features: ring,
hydrogen bond acceptor and donor, a hydrophobic group, ionizable groups). Similarity was
ranked by the values of TanimotoCombo coefficient [TCcombo], which is a sum of Tanimoto
coefficient for shape and color). Additionally, the area under the curve (AUC) and enrichment
factor values were calculated to infer the ability to discriminate active from inactive
compounds. Similarity models were also evaluated by the so-called confusion matrix, which
considers the number of true positives and negatives [TP and TR] and false positives and
negatives [FP and FR] classifications, together with the true positive and true negative rates
(TPR and TNR, respectively) themselves, accuracy (sum of true positive and true negative
classifications divided by total number of dataset compounds), F1 score and Matthews
correlation coefficient (MCC) as illustrated by equations 1 and 2, respectively [43].
ꢆꢇꢈ
ꢀ1 ꢁꢂꢃꢄꢅ = ꢆꢇꢈ ꢉꢊꢈꢉꢊꢋ
(Eq. 1)
(Eq. 2)
ꢇꢈ ꢎ ꢇꢋꢏꢊꢈ ꢎ ꢊꢋ
ꢌꢍꢍ = ꢐ ꢇꢈꢉꢊꢈ
ꢑ
ꢒ ꢑ
ꢒ ꢑ
ꢒ
ꢇꢈꢉꢊꢋ ꢇꢋꢉꢊꢈ ꢑꢇꢋꢉꢊꢋꢒ
2.3.6 External blind validation of QSAR and similarity models
At final of molecular modeling studies, a second external validation was carried using
compounds 10-15 for QSAR models and for ROCS model, denominated blind set. Figure 1
presents a flowchart of current work highlighting experimental approaches in blue squares
and computational ones in red squares.

13
ACCEPTED MANUSCRIPT
Figure 1. Experimental and computational workflow. Structure and pMIC values for C.
neoformans inhibition of compounds a-m [20].
3. Results and discussion
3.1 Chemistry
The hydrazine-thiazole derivatives 1-15 (Figure 2) were synthesized using a classical method
as previously reported [31] and briefly described below.

14
ACCEPTED MANUSCRIPT
Figure 2 - Chemical structure of the hydrazine-thiazoles synthesized.
The ketones 16a and 17a were prepared by crossed aldol condensation between acetone and
4-hydroxy-3-methoxybenzaldehyde or cyclohexanone and benzaldehyde in acidic and basic
conditions, respectively. Reaction of the ketones or the aldehydes with thiosemicarbazide
under acidic conditions, gave the corresponding thiosemicarbazones 16b-22b, which were
cyclized with α-bromoacetophenone to yield the thiazole heterocycles 1-15 (Scheme 1).

15
ACCEPTED MANUSCRIPT
Scheme 1 - General scheme for the preparation of hydrazine-thiazoles 1-15
All thiosemicarbazones had already been described and their melting point and spectroscopic
data are in accordance with the reported values in literature. The structures of the novel
hydrazine-thiazoles 1-15 were confirmed by spectroscopic methods and high-resolution mass
spectra. The 1H and 13C NMR data confirmed the formation of the thiazole moiety by presence
of a signal at around 7.0 ppm and 104.0 ppm, which corresponds to the methine proton and
carbon, respectively, of the thiazole ring. The hydrazine-thiazoles 1-3 were obtained as 1:1
1
mixture of E and Z isomers (C=N double bond), determined according to H NMR signal
integration of N-H at ~ 13.8 and 12.4 ppm. According to Tenchiu et al. (2009), the signal
assigned to N-H “appears at 9-12 ppm for the E-form, and 14-15 ppm for the Z-form” [49]. In
1
all other cases, (E)-isomers were obtained predominantly, as evidenced by H NMR
spectroscopic data and compared with similar structures [50].
3.2 Biological assays
All the hydrazine-thiazoles and the thiosemicarbazones 16b, 18b and 21b were tested for in
vitro antifungal activity against seven clinically important fungal species, namely Candida
albicans, C. krusei, C. parapsilosis, C. tropicalis, Cryptococcus neoformans, C. gatti and

16
ACCEPTED MANUSCRIPT
Paracoccidioides brasiliensis. Some thiosemicarbazones were included for comparison
purposes. The minimum inhibitory concentration (MIC) values are shown in Table 1.
Table 1 – Minimum inhibitory concentration (MIC) in µM of synthesized compounds
C.
C.
C.
C.
C.
C.
P.
Compound
albicans krusei parapsilosis tropicalis gatti
neoformans brasiliensis
>250
3.9
>250
3.9
>250
15.6
15.6
1.9
>250
3.9
>250
3.9
>250
3.9
>250
250
18b
1
3.9
7.8
7.8
1.9
3.9
62.5
250
2
0.9
1.9
3.9
0.45
>250
1.9
0.9
3
>250
3.9
>250
5.8
>250
3.9
>250
31.2
31.2
250
>250
250
250
250
ND
>250
1.9
>250
125
21b
4
3.9
7.8
3.9
0.9
1.9
250
5
1.9
2.9
3.9
0.9
1.9
125
6
>250
250
250
250
ND
1.5
>250
250
250
250
ND
>250
250
250
250
ND
>250
250
250
125
0.76
0.75
3.9
>250
250
250
250
1.5
>250
250
16b
7
ND
8
62.5
>250
>250
250
9
10
1.5
24.2
125
>250
>250
>250
3.2
5.9
0.75
3.9
11
31.2
>250
>250
>250
1.6
15.6
>250
>250
>250
104.4
250
>250
>250
>250
6.5
12
>250
81.2
>250
12.7
>250
20.3
>250
6.2
>250
>250
>250
ND
13
14
15
Fluconazole
Amphotericin
B
0.3
0.7
0.5
0.5
1.2
0.08
0.18
0.09
0.04
0.06
0.06
0.18
0.18
0.09
Itraconazole
ND = not determined
From the obtained results, it can be concluded that the presence of the thiazole ring was
essential for the activity since the corresponding thiosemicarbazones 16b, 18b and 21b that
lacks such ring were inactive. It is possible to infer also that the presence of a hydrophobic
aliphatic chain plays a very important role in the antifungal activity. Since, the most
promising compounds 1-6 and 10-12 have an aliphatic substituent in the hydrazine moiety. In
general, the presence of the chlorine substituent in aromatic ring (R3 in Scheme 1) was shown
to increase activity with regard to all fungal species tested (1, 2 vs 3 and 4, 5 vs 6), except for
C. tropicalis and P. brasiliensis. The active hydrazine-thiazoles showed a broad spectrum of

17
ACCEPTED MANUSCRIPT
action against several pathogenic fungi, some of them being equally or more active than the
positive controls fluconazole and amphotericin B. It is worth mentioning that compound 3
showed more activity than fluconazole against all fungal strains. Compounds 10 and 11 were
the most potent against C. neoformans and C. gattii and are therefore very promising
antifungal candidates. The antifungal activity of compound 10 against various clinical fungal
isolates of Candida spp. has previously been described by Carradori et al. (MIC = 2-32
µg/mL) [26] and therefore this compound was not tested against Candida spp. in the present
work. None of the compounds was sufficiently active against P. brasiliensis.
The results obtained in our studies are in accordance with those obtained by other authors that
also reported the anti-Candida activity of hydrazine-thiazoles with hydrophobic features
[26,51,52]. However, to the best of our knowledge, the activity of this class of molecules
against Cryptococcus species is still very little studied. The mechanism of action of this class
of compounds has not yet been fully elucidated. However, recently, our research group has
shown that the activity of these thiazoles can be related to an increase in intracellular reactive
oxygen species (ROS) such as superoxide, causing oxidative stress in the fungal cells [53].
In addition, the cytotoxicity of the active compounds was evaluated in human embryonic
kidney (HEK-293) cells and the results are shown in Table 2. All compounds displayed low
cytotoxicity toward human cells with IC50 values higher than at least 50 µM. These results
demonstrate the enormous potential of these hydrazine-thiazoles with regard to both
antifungal activity and selectivity.
Tabela 2 - Cytotoxicity of compounds on HEK cells
IC50
(µM)
% viability at highest
concentration tested
Compounds
> 250*
207.00 ± 12.89
237.86 ± 9.71
> 50*
52.41 ± 6.58
-
1
2
-
3
94.86 ± 6.96
94.96 ± 7.22
76.23 ± 7.40
66.10 ± 6.00
-
5
> 50*
6
> 250*
10
11
> 250*
184.69 ± 13.72
> 50*
12
87.65 ± 7.83
14
78.62 ± 10.93
> 250*
Amphotericin B
Fluconazole
Itraconazole
-
99.21 ± 6.04
> 250*
90.76 ± 6.47
*Highest concentration tested based on the solubility of the compound in the assay medium. Data are
presented as mean ± standard deviation obtained from three independent experiments performed in
triplicate. Itraconazole and Fluconazole data were obtained from two experiments.

18
ACCEPTED MANUSCRIPT
3.3 Molecular modeling studies
3.3.1 QSAR Studies
Initially, the dataset division into training and test set was achieved its main goal: to distribute
training and test set compounds homogeneously according to its biological, structural and
chemical spaces. Supplementary Figure S1 illustrates that both training and test set
compounds are distributed in the same proportion of complete dataset compounds in each
observed cluster of all three analyzed spaces. Therefore, the selected test set compounds were
suitable to perform external validations procedures. The molecular alignment was also
achieved with success since all hydrazine-thiazole scaffold and benzene rings are perfectly
aligned (Supplementary Figure S2).
Using selected training set compounds, the first 16 HQSAR models were constructed using C.
neoformans pMIC values by varying the fragment distinction and fixing the fragment size
parameters. The two most robust HQSAR models considered atoms, connections, and
presence of H-bond acceptors/donors (q2 = 0.792) and atoms, bonds, connections and
presence of H-bond acceptors/donors (q2 = 0.757). Then, these two models were employed to
fragment size variation originating 18 other models. The fragment size variation indicated that
larger fragments have better correlation with biological activity (in this case, the most robust
model has a q2 equal to 0.877). All HQSAR models information is available at Supplementary
Table S1 and the five most robust models which were submitted to external validations are
shown in Table 3.
Table 3 - Five most robust HQSAR models using fragment size equals to 4 to 7 atoms.
q2
SEV
r2
SEE HL
PC
4
4
4
4
No.
Fsize
8_11
10_13
7_10
8_11
10_13
Fdist
23
25
31
32
34
A/C/DA
A/C/DA
A/B/C/DA
A/B/C/DA
A/B/C/DA
0.854
0.828
0.860
0.877
0.834
0.439 0.966 0.210 61
0.476 0.972 0.193 61
0.429 0.974 0.187 97
0.403 0.975 0.181 59
0.468 0.963 0.221 83
4
Fsize: fragment size (number of atoms); Fdist: fragment distinction; q2: leave-one-out internal validation
coefficient; SEV: standard error of validation; r2: calibration coefficient; SEE: standard error of
estimation; HL: hologram length (bins); PC: number of PLS principal components.
Then, a standard CoMFA model was generated obtaining in an unsatisfactory result (q2 =
0.531). Therefore, we applied the region focusing strategy in order to improve the robustness
of CoMFA models. From all 28 generated CoMFA models, eighteen models presented
acceptable robustness (q2 > 0.6) and the five most robust were generated using the original
distance between grid points (2Å). The five most robust models (Table 4) indicate that steric
effects have higher contribution than electrostatic one, indicating that van der Waals effects
could be slightly more important to antifungal activity of studied series. This result
corroborates SAR studies presented in this work and presented in previous work reporting

19
ACCEPTED MANUSCRIPT
part of dataset compounds [20]. Models reported in Table 4 were also submitted to external
validations.
Table 4 - Five most robust CoMFA models using region focusing technique.
Model
W
d
1
1
1
1
1
q2
SEV
0.604
0.584
0.579
0.592
0.601
PC
4
4
4
4
r2
SEE
S
E
6
9
12
15
18
0.5
0.7
0.9
1.1
1.2
0.741
0.758
0.762
0.752
0.744
0.979
0.979
0.980
0.980
0.979
0.174 0.521 0.479
0.173 0.528 0.472
0.168 0.553 0.447
0.169 0.589 0.411
0.173 0.611 0.389
4
w: weight factor based on standard deviation; d: distance factor applied on distance of grid points; q2:
leave-one-out internal validation coefficient; SEV: standard error of validation; r2: calibration coefficient;
SEE: standard error of estimation; S: fraction of steric contribution to model; E: fraction of electrostatic
contribution to model; PC: number of PLS principal components
After HQSAR and CoMFA model construction, 31 standard CoMSIA models were generated
by combining similarities indexes (steric, electrostatic, hydrophobic, H-bond acceptor and
donor). In this step, only one model constructed using H-bond donor (D) similarity index
presented acceptable q2 value (equal to 0.702). In addition, a second most robust model
(electrostatic/donor, q2 = 0.656) was also selected to region focusing technique. In total, 85
CoMSIA models were constructed and compared according to their robustness
(Supplementary Table S3 and S4). From this analysis, five most robust CoMSIA models were
selected and submitted to external validations (Table 5). All top five CoMSIA models were
generated using both H-bond donor (D) and electrostatic indexes and, in all cases, D
presented a minor contribution to models, indicating the importance of electrostatic
interactions to explain the antifungal activity.
Table 5 - Five most robust CoMSIA models using region focusing technique.
q2
SEV
PC
4
4
4
4
r2
SEE
0.298
0.275
0.278
0.293
0.301
D
E
Model
W
d
66
69
72
75
78
0.5 1.5 0.843 0.471
0.7 1.5 0.866 0.435
0.9 1.5 0.883 0.406
1.1 1.5 0.876 0.419
1.2 1.5 0.869 0.429
0.937
0.947
0.945
0.939
0.936
0.193
0.181
0.185
0.184
0.182
0.807
0.819
0.815
0.816
0.818
3
w: weight factor based on standard deviation; d: distance factor applied on distance of grid points; q2:
leave-one-out internal validation coefficient; SEV: standard error of validation; r2: calibration coefficient;
SEE: standard error of estimation; D: fraction of H-bond donor contribution to model; E: fraction of
electrostatic contribution to model; PC: number of PLS principal components.
After generation and comparison of robustness of 2D- and 3D- QSAR models, top five most
robust models were submitted to external validations in order to select the most predictive
model of each technique to continue QSAR studies. Then, models HQSAR 23, CoMFA 6 and

20
ACCEPTED MANUSCRIPT
CoMSIA 75 showed higher Q2(F1), Q2(F2), Q2(F3), r2 metrics and CCC values than other
m
models as well as lower error metrics values (RMSEP and MAE) (Table 6).
Table 6 - Comparison of external validation metric for top 5 most robust models generated with each
QSAR technique.
HQSAR
CoMFA
CoMSIA
Model
25
31
32
34
9
12
15
18
66
69
72
78
23
6
75
Q2
0.930 0.928 0.927 0.938
0.925 0.923 0.922 0.934
0.934 0.932 0.931 0.941
0.911 0.889 0.853 0.965
0.916 0.938 0.902 0.933
0.005 0.049 0.049 0.032
0.855 0.855 0.854 0.853 0.756 0.823 0.865
0.845 0.845 0.844 0.843 0.740 0.811 0.856
0.863 0.863 0.862 0.861 0.770 0.833 0.872
0.697 0.681 0.632 0.620 0.725 0.789 0.824
0.821 0.810 0.797 0.790 0.839 0.877 0.905
0.124 0.128 0.164 0.170 0.114 0.088 0.080
0.880
0.872
0.887
0.827
0.913
0.087
0.957
0.954
0.959
0.945
0.931
0.014
0.871
0.863
0.879
0.714
0.841
0.127
0.880
0.872
0.887
0.832
0.919
0.087
(F1)
Q2
(F2)
Q2
(F3)
r2
m
r2 '
m
∆r2
m
AVGr
2
0.913 0.913 0.878 0.949
0.962 0.960 0.957 0.967
0.759 0.745 0.714 0.705 0.782 0.833 0.865
0.916 0.914 0.913 0.911 0.874 0.908 0.929
0.870
0.937
0.938
0.977
0.777
0.926
0.875
0.937
m
CCC
RMSE
P
0.259 0.262 0.265 0.244
0.219 0.232 0.234 0.201
0.373 0.374 0.375 0.375 0.484 0.412 0.360
0.297 0.307 0.317 0.321 0.393 0.318 0.269
0.339
0.232
0.203
0.181
0.352
0.237
0.340
0.241
MAE
Q2(F1): predictive R2 ; Q2(F2): Schüürmann et al. q2; Q2(F3): Consonni et al. q2; r2 : Roy parameter to access predictive power; r2 ':
m
m
predictive power calculate with inverted axis; ∆r2 : r2 - r2 '; AVGr2 : average r2 value; CCC: concordance correlation coefficient;
m
m
m
m
m
RMSEP: root mean squared error of prediction for test set; MAE: mean absolute error
.
Then, these metrics reflect the predictive power of model and its capacity to predict the
biological activity of compounds out of pMIC range employed in this study. Therefore, the
most predictive models were submitted to leave-N-out cross-validation and progressive-
scrambling as last validations. Those final validations indicated that both models were not
obtained by chance (scrambled models q2 lower than original model standard error of
validation higher than original model), and models are consistent under training set variations
(all leave-N-out validations coefficients were higher than 0.6) and did not show residual
values higher than 1 logarithmic unit (Figure 3).

21
ACCEPTED MANUSCRIPT
Figure 3 - Experimental versus predicted pMIC values (A, training set compounds are
presented in white circles and test set compounds are presented in black square), leave-N-out
internal validation (B) and progressive-scrambling (C, grey square represents original model
and black circles represent scrambled models) and residual values for test set compounds.
Finally, validated models were employed to predict pMIC values for a blind set of compounds
(Table 7). A comparison between residual values of test set and blind set indicated that 2D
QSAR model outperform 3D ones in prediction of pMIC for compounds 10-15. We also
estimated the applicability domain of generated QSAR models by analyzing the percentage of
missing fragments (for HQSAR model) and the percentage of extrapolated variables (for
CoMFA and COMSIA models) of external validations set compounds in relation to training
set compounds. This analysis demonstrate how test and blind set compounds are similar to
training set ones according to their respective chemical spaces used to construct QSAR
models: structural similarity for HQSAR, steric and electrostatic fields similarity for CoMFA,
and electrostatic and H-bond donor similarity for CoMSIA. We concluded that high errors of
prediction for blind set compounds (in special for compounds 13-15) could be associated with
the fact of blind set compounds are far from applicability domain than test set ones for
CoMFA and CoMSIA models.

22
ACCEPTED MANUSCRIPT
Table 7 - Internal and external (using blind set) validations for top10 ROCS models.
HQSAR
CoMFA
% miss.d pred.b res.c % miss.e pred.b
CoMSIA
res.c
%
cpds exp.a pred.b
res.c
miss.e
test set
4.807 4.969 -0.162
0.00
0.00
0.00
0.00
5.456 -0.649
5.832 0.214
5.822 -0.101
3.731 -0.129
blind set
3.25
25.26
2.28
5.459 -0.652
6.27
1a
1d
4
6.046 5.716
0.330
6.199 -0.154 10.36
5.721 5.800 -0.079
5.646
0.075
5.06
2.41
3.602 3.447
0.155
3.83
3.685 -0.083
7
5.824 5.294
6.125 5.284
5.409 5.177
0.530
0.841
0.232
0.00
1.00
9.00
8.00
5.00
0.00
5.371 0.453
5.715 0.410
5.768 -0.359
5.172 -1.570
5.171 -1.384
14.60
10.99
25.05
31.00
30.02
22.31
6.282 -0.458 28.19
6.398 -0.273 26.27
6.336 -0.927 29.64
6.188 -2.586 15.90
4.270 -0.482 34.94
10
11
12
13
14
15
3.602 4.438 -0.836
3.788 4.901 -1.113
3.602 3.931 -0.329
4.847 -1.245
6.282 -2.463 32.53
aexperimental pMIC values; predicted pMIC values; residual of prediction (a-b); percentage of missing
b
c
d
fragments (HQSAR) of external validation compounds in relation to training set ones; epercentage of
extrapolated terms from grid points (CoMFA and CoMSIA) of external validation compounds in relation to
training set ones.
From HQSAR contribution maps (Figure 4), the hydrazine-thiazole group and the benzene
attached to it as well as an aliphatic substituent of N-R region presented green, blue-green and
yellow contributions, indicating that these groups are important to antifungal activity. In
contrast, an aromatic ring of compound f located at the hydrophobic region is colored in red
and orange, indicating that aromatic substituent at this region would contribute negatively to
antifungal activity.

23
ACCEPTED MANUSCRIPT
Figure 4 - Contribution maps generated with the best HQSAR model.
From CoMFA maps (Figure 5), both substituents at hydrazine group of most active
compounds (c and 2) are located at steric favored regions and pyridine ring of compound e is
located at a steric impeachment region. Interestingly, the N atom of pyridine ring of
compound e could form electrostatic interactions compensating the disfavored bulky region of
this compound. Also, there is a very small green polyhedral located at methoxy substituent of
benzene indicating that bulky groups at these regions could be slightly more active than
others. There is a negatively charged favored region located near the three N atoms of thiazole
ring and hydrazine group indicating that electron lone pairs could form an electrostatic
interaction with its molecular target. Furthermore, there is a positively charged region located
at hydroxyl group of hydrazine substituent of compound c.

24
ACCEPTED MANUSCRIPT
Figure 5 - Contour maps generated with best CoMFA model
Electrostatic CoMSIA contour maps (Figure 6) corroborates the CoMFA results, with one
exception, the negative similarity region is located only at N atom of hydrazine group. Then,
this finding plus the higher contribution of electrostatic similarity index for CoMSIA model
could indicate the major contribution of hydrazine group to electrostatic interactions than
thiazole ring. Finally, NH of hydrazine group is located at a favored H-bond donor region
indicating the importance of an H-bond to biological activity.

25
ACCEPTED MANUSCRIPT
Figure 6 - Contour maps generated with best CoMSIA model.
3.3.2. Molecular Similarity Studies
Initially, five active compounds were employed to construct 15 models based on 3D
molecular shape, chemical features and the sum of both. The AUC values ranged from 0.736
to 0.987, with no statistical difference between models: all confidence levels of AUC values
were superimposed. Then, 74 new models were constructed by applying weight to each of the
chemical features, by excluding chemical features or by the combination of both strategies. In
this step, we did not use compound i due to lowest AUC values presented by models where it
was employed as a template. This strategy allowed us to evaluate the importance of the
individual chemical groups to the classification of compounds in active and inactive. All the
89 generated similarity models are reported in Supplementary Table S5.
All 89 generated similarity models were reported in Supplementary Table S5. Then, we
selected 10 models with highest AUC values to validate it by a confusion matrix construction
and calculation of classification metrics. In this step, we calculated the similarity between all
dataset compounds with a query model and defined a cutoff value of TCcombo (to classify
compounds in active or inactive) according to the first decoy found in the ranking. Table 8
displays calculated results.

26
ACCEPTED MANUSCRIPT
Table 8 - Internal and external (using blind set) validations for top10 ROCS models.
Models
72
53
24
14
52
70
16
84
60
18
76
0.988
0.982
1.000
95.58
0.911
1.000
0.974
0.975
0.637
0.588
1.000
0.000
0.987 0.987 0.986 0.986 0.986
0.981 0.982 0.977 0.981 0.979
1.000 1.000 1.000 1.000 1.000
118.92 97.32 95.19 94.97 88.81
1.172 0.902 1.144 0.911 1.135
1.000 1.000 1.000 1.000 1.000
0.956 0.974 0.956 0.974 0.956
0.956 0.975 0.956 0.975 0.956
0.530 0.637 0.530 0.637 0.530
0.455 0.588 0.455 0.588 0.455
1.000 1.000 1.000 1.000 1.000
0.000 0.000 0.000 0.000 0.000
0.984 0.984
0.971 0.947
1.000 1.000
71.69 62.67
1.133 1.161
1.000 1.000
0.956 0.959
0.956 0.960
0.530 0.548
0.455 0.476
1.000 1.000
0.000 0.250
AUC
0.988
0.975
1.000
44.29
1.250
1.000
0.974
0.975
0.637
0.588
1.000
0.750
0.985 0.984
0.977 0.971
1.000 1.000
82.09 82.87
1.212 1.397
1.000 1.000
0.970 0.967
0.971 0.967
0.611 0.588
0.556 0.526
1.000 0.500
0.750 0.750
- 95% c.l.a
+ 95% c.l.b
EF 0.5% c
TCcombo cut-off
TPR
TNR
ACC
MCC
F1
TPRext
TNRext
ACCext
MCCext
F1ext
0.333 0.833 0.333 0.333 0.333 0.333 0.333 0.833 0.667 0.333 0.500
0.000
0.500
0.000 0.000 0.000 0.000 0.000
0.500 0.500 0.500 0.500 0.500
0.000 0.158
0.500 0.571
0.612
0.800
0.612 0.250
0.800 0.500
a, b = lower and upper confidence levels values of AUC calculated with bootstrap validation, respectively; c = enrichment factor
(actives found / decoys found) at 0.5% of entire database; ext = metrics calculated with blind external validation set
Comparison of values of the confusion matrix shows no statistical differences among the ten
most robust models. Then, the external validation using compounds 10-15 was fundamental to
distinguish models due its capacity of to classify compounds according to active and inactive.
Only models 53, 84 and 60 (Figure 8B-D) presented acceptable values of accuracy (> 60%),
while all other seven models could be considered worse than a random model, due to low
ACC values for external validation set (< 0.5). Another important observed characteristic of
those top three models is that they have the highest TC-combo values defined as cutoff. This
result indicates that active compounds of external validation set are more similar to model
template. From top three models, model 84 is the first one to find all active molecules at ROC
curve (Figure 8A). Altogether, these findings indicate the importance of hydrophobic feature
as hydrozone substituent. For all selected models, the benzene and thiazole rings as well as
hydrazone moiety has same features, indicating that these groups are important to antifungal
activity.
Among the best three models, model 53 has a methoxyl substituent at the benzene ring, which
when exploited during the ROCS model generation lead to more discriminant models, if an H-
bond acceptor feature was included. This finding could indicate that this region could not
directly interact with possible molecular targets and that only its shape influences the
biological activity.

27
ACCEPTED MANUSCRIPT
Figure 8 - (A) ROC curve of 3 best models and visual representation of model 53 (B),
model 60 (C) and model 84 (D). Molecular features of models are presented as green spheres
(rings), red spheres (H-bond acceptors), blue spheres (H-bond donors) and yellow/white
spheres (hydrophobic groups).
From all molecular modeling studies (2D- and 3D-QSAR models as well as similarity
models), several features were observed and showed importance to antifungal activity. In
other words, those groups showed mathematical importance to pMIC prediction (or
classification) and could be related to interactions with possible molecular targets. A brief
summary of observed structure-activity relationship from molecular modeling studies is
shown in Figure 9.
Figure 9 - Main observed features from HQSAR, CoMFA, CoMSIA and ROCS models.

28
ACCEPTED MANUSCRIPT
4. Conclusions
Evaluation of in vitro antifungal activity of novel hydrazine thiazoles revealed that several of
them showed potent activity against clinically important Candida and Cryptococcus species,
with MIC values ranging from low micromolar to nanomolar. These results demonstrate the
significant potential of this class of compounds as antifungal agents. In addition, the active
compounds showed low cytotoxicity to human embryonic kidney cells.
All generated QSAR and similarity models were robust and have high predictive power in
performed validations. Furthermore, all models were employed to provide physicochemical
interpretations, which corroborated experimental SAR studies and, therefore, they can be
employed in the design of new antifungal agents. Further synthesis followed by molecular
modeling studies will be carried out to increase the antifungal potency of hydrazine-thiazole
derivatives. In this sense, new QSAR models can be generated using compounds 10-15 in
training set, as well as, planned derivatives in order to increase applicability domain of
proposed models. Both generated 2D- and 3D-QSAR models presented satisfactory internal
and external validations parameters and further studies could increase the accuracy of models
and increase the range of prediction for new compounds.
Conflict of interest
The authors state that they have no conflict of interest.
Acknowledgements
The authors thank the Fundação de Amparo à Pesquisa de Minas Gerais (FAPEMIG),
Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) and Coordenação
de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) for financial support. The authors
thank also Centro de Pesquisas René Rachou (FIOCRUZ) for providing high-resolution mass-
spectra and OpenEye Scientific Software for OMEGA and ROCS academic licenses.
References
[1] D.A. Enoch, H. Yang, S.H. Aliyu, C. Micallef, The changing epidemiology of invasive
fungal infections, Methods Mol. Biol. 1508 (2017) 17-65. Doi: 10.1007/978-1-4939-6515-1_2
[2] Ravikant, T. Kaur, S.Gupte, M. Kaur, A review on emerging fungal infections and their
significance, J. Bacteriol. Mycol. 1 (2015) 1-3. Doi: 10.15406/jbmoa.2015.01.00009
[3] L.C. Whitney, T. Bicanic, Treatment principles for Candida and Cryptococcus, Cold
Spring Harb. Perspect. Med. 10 (2014) a024158. Doi: 10.1101/cshperspect.a024158