In vitro and in silico anti-dengue activity of compounds obtained from Psidium guajava through bioprospecting

Background For decades, bioprospecting has proven to be useful for the identification of compounds with pharmacological potential. Considering the great diversity of Colombian plants and the serious worldwide public health problem of dengue—a disease caused by the dengue virus (DENV)—in the present study, we evaluated the anti-DENV effects of 12 ethanolic extracts derived from plants collected in the Colombian Caribbean coast, and 5 fractions and 5 compounds derived from Psidium guajava. Methods The cytotoxicity and antiviral effect of 12 ethanolic extracts derived from plants collected in the Colombian Caribbean coast was evaluated in epithelial VERO cells. Five fractions were obtained by open column chromatography from the ethanolic extract with the highest selectivity index (SI) (derived from P. guajava, SI: 128.2). From the fraction with the highest selectivity (Pg-YP-I-22C, SI: 35.5), five compounds were identified by one- and two-dimensional nuclear magnetic resonance spectroscopy. The antiviral effect in vitro of the fractions and compounds was evaluated by different experimental strategies (Pre- and post-treatment) using non-toxic concentrations calculated by MTT method. The DENV inhibition was evaluated by plate focus assay. The results were analyzed by means of statistical analysis using Student’s t-test. Finally the antiviral effect in Silico was evaluated by molecular docking. Results In vitro evaluation of these compounds showed that three of them (gallic acid, quercetin, and catechin) were promising antivirals as they inhibit the production of infectious viral particles via different experimental strategies, with the best antiviral being catechin (100% inhibition with a pre-treatment strategy and 91.8% with a post-treatment strategy). When testing the interactions of these compounds with the viral envelope protein in silico by docking, only naringin and hesperidin had better scores than the theoretical threshold of − 7.0 kcal/mol (− 8.0 kcal/mol and − 8.2 kcal/mol, respectively). All ligands tested except gallic acid showed higher affinity to the NS5 protein than the theoretical threshold. Conclusion Even though bioprospecting has recently been replaced by more targeted tools for identifying compounds with pharmacological potential, our results show it is still useful for this purpose. Additionally, combining in vitro and in silico evaluations allowed us to identify promising antivirals as well as their possible mechanisms of action.


Introduction
Dengue is a growing public health problem worldwide, mainly in tropical and subtropical regions [1]. In the last five decades, on the American continent, the incidence of dengue has increased 30 fold and is the cause of approximately 390 million infections per year, of which 96 million have clinical manifestations [2]. In 2013, in America, the largest number of cases in the history of the disease was reported, with a total of 2.3 million cases, an alarming figure [3,4]. In most cases, dengue presents as an asymptomatic disease, however, it can presented with a wide range of clinical manifestations, including fever, headache, pain in various parts of the body, and prostration, among others. Few patients present with serious life-threatening manifestations. This wide clinical variety permits classified dengue in three groups: asymptomatic patients, symptomatic patients (with or without alarming symptoms) and patients with severe dengue (patients with hemorrhage and/or hypovolemic shock)") [5].
The etiological agent that causes this disease is the dengue virus (DENV), a member of the Flaviviridae family, which belongs to the arboviruses (viruses transmitted by arthropods). This virus is classified into four serotypes (DENV 1-4) according to genetic and antigenic differences [6]. Although any serotype is equally able to cause dengue, serotype differences have been postulated to lead to differences in pathogenesis [7] such as the case for DENV-2 which have been related with sever dengue [8]. The DENV genome consists of a strand of positive-sense Ribonucleic Acid (RNA) of approximately 1 kb with a m7GpppAmp cap at its 5′ end and no poly (A) tail at its 3′-end [9]. Moreover, it has a single reading frame that codes for three structural proteins (C, PrM, and E) and seven nonstructural proteins (NS1, NS2a, NS2b, NS3, NS4a, NS4b, and NS5), which are mainly involved in viral replication [10,11].
The replication cycle of DENV begins with the E protein binding to receptors on the cell membrane. After this binding, there is a clathrin-mediated endocytosis followed by the formation of an endosome leading to the pHdependent fusion (viral envelope/endosome). The acidic pH of the endosome favors the release of the viral RNA in the cytoplasm to be transcribed and translated into ribosomes associated to the endoplasmic reticulum, resulting in a single viral polyprotein which is cleaved by cellular and viral proteases. The replication complex (formed by RNA, NS5, other nonstructural proteins and cellular factors) is formed in association with intracellular membranes. Finally, the assembly of viral proteins with new genomes occurs in the lumen of the endoplasmic reticulum, followed by the passage of the new virions through the Golgi apparatus (where the maturation viral process occurs) to be released the virions by exocytosis [12]. Although all steps of the viral replication cycle are likely to be inhibited, most studies have focused on the evaluation of compounds that inhibit the attachment and entry of virus into the cell, among them are heparin [13] and sulfated polysaccharides [13]. Moreover, other compounds are able to inhibit the viral genome replication by blocking the synthesis of nucleoside triphosphates. Ribavirin and mycophenolic acid [14] are two good examples of such inhibitors. Finally, other compounds, including castanospermine [15] and the Lovastatin [16] inhibit steps in the replication cycle after entry and replication of DENV, possibly affecting the assembly process.
Despite the large number of possible antiviral candidates [17], to date, only a few have been tested in clinical trials, such as balapiravir, chloroquine, lovastatin, prednisolone, and celgosivir [18]. However, none of these compounds is being used as an effective anti-dengue therapy. Thus, there remains an important need to identify effective and tolerable medications for treatment of DENV-infected patients both in the early phase, to prevent progression to fatal outcomes, and to minimize deaths after patients develop severe complications [19]. For this reason, the agenda of research priorities proposed by the World Health Organization proposed, a decade ago, that included searching for antivirals (either second-use drugs or natural product derivatives), is still valid today [20]. Moreover, other of the main reasons for the high incidence of the DENV worldwide is that thus far, only one licensed vaccine exists for use in a few countries [21]. This vaccine was produced by Sanofi Pasteur (CYD-TDV) is composed of four attenuated vaccines (CYD-1-4) is a life recombinant vaccine, based on a yellow fever vaccine 17D (YFV 17D) backbone [22].
Plants have been used as medicinal sources to treat many diseases for thousands of years. Ancestral communities could empirically identify plants to fight infections, passing many of these findings from generation to generation until today [23]. Ethnobotany has proven useful for preserving such knowledge in communities that use plants to treat diseases has encouraged bioprospecting studies, which allow the identification of compounds with pharmacological potential [24]. However, is important highlight that in some cases the communities, without scientific knowledge, use plants included in the IUCN Red List of Threatened Species [25] to treat diseases, and for these reason is important encourage to preserve, protect and promote the traditional knowledge but with scientific support.
Due to the diversity of chemical compounds present in plants, they are an important source of pharmacological candidates. Several studies have identified potential plant derived candidates compounds as a potential of new drugs candidates that inhibit the activity of viruses such as herpes virus [26], hepatitis C virus, [27], human immunodeficiency virus type I [28], rotavirus [29], influenza virus, [30], chikungunya virus [31], and DENV [32]. Specifically, for DENV, the antiviral effects of some compounds has been demonstrated, inhibiting both, infectivity and/or viral spread in vitro and in vivo, such as glabranine (derived from Tephrosia madrensis) [33], panduratin (derived from Boesenbergia rotunda L.) [34], and castanospermine (derived from Castanospermum australe) [15]. Recently, inhibitory activity against DENV-2 infection has been identified in ethanolic extracts of Colombian plants such as Cassia grandis and Tabernaemontana cymosa (T. cymosa), with 99.9% inhibition observed for the T. cymosa extract [35]. Moreover, several compounds derived from Mammea americana (coumarins) and T. cymosa (lupeol and voacangine) that can inhibit infection in vitro at percentages greater than 50% have been reported [36].
Given the great variety of Colombian plants, the present study evaluated the anti-DENV effect of 12 ethanolic extracts obtained from selected plants based on an ethnobotanical survey conducted in the city of Cartagena (in the Colombian Caribbean coast). The ethanolic extract with the highest selectivity (derived from Psidium guajava) was fractionated, and the anti-DENV effect of each fraction was tested. Moreover the results of in vitro assay were contrasted with the in silico assays to postulate promising antivirals.

Plant selection
Vegetal material was selected based on data obtained from an ethnobotanical survey conducted in the city of Cartagena (Colombia) in 2009 (unpublished results) and from a literature search on plant extracts with reported antiviral activity against viruses that cause febrile illness. In total twelve plants were included: Ambrosia cumanensis Kunt (A. cumanensis), Cavanillesia platanifolia Bonpl (C. platanifolia), Chenopodium ambrosioides L.

Obtaining extracts and fractions
Weighed amounts (1000 g) of each plant material were collected for testing (phytochemical and biological examinations) and for identification at the herbarium of the Universidad de Antioquia (HUA) (Medellín, Colombia) and at the herbarium of the Universidad Nacional de Colombia (COL) (Bogotá, Colombia). Moreover, the material was identified by authorized personnel at the Jardín Botanico Guillermo Piñeres in Cartagena (JBC), Colombia (voucher number JBC 1209). The collected plant material was dried at room temperature, weighed, ground, and macerated with 95% ethanol for 72 h. Each sample was then filtered and concentrated in a rotary evaporator. Each dried extract was suspended in a mixture of ethanoldistilled water and subjected to liquid-liquid partitioning with solvents of increasing polarity in the following order: dichloromethane, ethyl acetate, and butanol, as shown in Fig. 1. For the preliminary phytochemical screening of each extract, identification tests were performed for different secondary metabolites according to the method described in [37].

Isolation and purification of the compounds present in the active fractions
The fractions were subjected to several column chromatographic procedures using silica gel and Sephadex and to normal-phase preparative thin-layer chromatography (PTLC), depending on their polarities, molecular size, and complexity. Final purification was performed using normal or reverse-phase high-performance liquid chromatography (HPLC) as necessary. The structures of the compounds present in the most active fraction of P. guajava (Pg-YP-I-22C) were elucidated using standard analytical methods, including melting point determination, co-chromatography with reference compounds, and 1 H nuclear magnetic resonance (NMR) and 13 C NMR spectroscopy techniques in one and two dimensions, such as the attached proton test (APT), distortionless enhancement of polarization transfer (DEPT), correlation spectroscopy (COSY), nuclear Overhauser effect spectroscopy (NOESY), heteronuclear multiplequantum correlation spectroscopy (HMQC), and heteronuclear multiple-bond correlation spectroscopy (HMBC).

Virus and cell maintenance
Epithelial VERO cells (Cercopithecus aethiops) were obtained from the American Type Culture Collection (ATCC), and C6/36HT cells (from Aedes albopictus mosquito larvae) were donated by Dr. Guadalupe Guzmán, Department of Virology, Pedro Kouri Institute (Havana, Cuba). C6/36HT cells were used to produce the viral stocks and the VERO cells were used to made the antiviral assays.
The cells and the virus strain were maintained according to the protocols described previously [38] . Briefly, the VERO cells were maintained in Dulbecco's Modified Eagle's medium (DMEM; Gibco®) supplemented with 2% fetal calf serum (FCS, Gibco®) and were incubated at 37°C in a 5% CO 2 atm. The C6/36HT cells were maintained in DMEM (Gibco®) supplemented with 10% FCS (Gibco®) at 34°C. The DENV 2/NG strain, which was used for all biological assays, was donated by Dr. Jorge Osorio, Department of Pathobiological Sciences, University of Wisconsin (Madison, WI, United States).

Determination of the selectivity index (SI)
The SI of each ethanolic extract, fraction, or compound was calculated from the relationship between the cytotoxic concentration 50% (CC 50 ) and effective concentration 50% (EC 50 ) [39]. The CC 50 was determined via the 3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide) (MTT, Sigma-Aldrich) method. For this, 2.5 × 10 4 VERO cells were seeded in 96-well plates for 24 h. Subsequently, serial dilutions of the ethanolic extracts (7.8 μg/mL to 1000 μg/mL), fractions (6.25 to 400 μg/mL), and characterized compounds (6.25 to 400 μg/mL) were performed and incubated with the cells for 48 h. After the incubation period, an MTT solution (0.5 mg/mL) was added to the cultures, which were then incubated for an additional 3 h at 37°C. Finally, dimethyl sulfoxide was added, and the absorbance was read at 450 nm in a Bio-Rad Benchmark® microplate reader. Cultures without extract, fractions, or compounds were processed as positive controls for viability. The CC 50 was calculated as the extract concentration that reduced cell viability by 50% using regression analysis (Probit software). Each experimental condition was tested in triplicate in two independent experiments (n = 6). To calculate the EC 50 , 2.5 × 10 4 VERO cells were seeded in 96-well plates for 24 h. At 24 h, serial dilutions of each ethanolic extract (7.8 μg/mL to 1000 μg/mL), fraction (6.25 to 400 μg/mL), and characterized compound (6.25 to 400 μg/mL) were mixed with the DENV-2/NG strain (at a multiplicity of infection (MOI = 1) and added to the cell monolayers for 2 h. At 2 h post-inoculation (hpi), the mixtures were removed, and the same serial dilutions of the extracts, fractions, or compounds were added again and incubated for 24 h. Finally, the supernatants were collected and stored at − 70°C until performing the multiwell plate focus assay [40].

Multiwell plate focus assay
A total of 2.5 × 10 4 cells/well were seeded in 96-well plates. The cells were inoculated the following day with serial dilutions (10 − 1 to 10 − 6 ) of the supernatants from the EC 50 experiments for 2 h. Next, the inoculum were removed, and 1.5% carboxymethylcellulose (Sigma-Aldrich) prepared in DMEM (Gibco®) supplemented with 2% FCS (Gibco®) was added, followed by incubation at 37°C in a 5% CO 2 atmosphere. On the third day, the cells were fixed with a methanol-acetone (1:1) solution for 10 min, washed three times with PBS, and permeabilized with 0.1% Triton X-100 for 30 min, after which nonspecific site blocking was performed with 10% FBS in PBS. Then, DENV anti-envelope monoclonal antibody (mAb) was diluted 1:500 with PBS containing 10% FCS (Gibco®) was added, followed by incubation for 1 h at 37°C. After washing with PBS, the plates were incubated for 30 min with an anti-mouse IgG secondary antibody conjugated to peroxidase. Finally, the reactions were stained with 3amino-9-ethylcarbazole (AEC, Sigma-Aldrich), and the number of foci in each well were counted.
Determining the antiviral effects of the fractions and compounds on virus cell entry A total of 2.5 × 10 4 VERO cells were seeded in 24-well plates for 24 h, and the fractions (100 μg/mL) or characterized compounds (50 μg/mL) were then added and incubated with the cells for an additional 24 h (pretreatment strategy) [16]. Subsequently, treatment was withdrawn, and a viral inoculum (DENV-2/NG at an MOI = 1) was added and incubated for 2 h. Subsequently, the virus was removed, and fresh medium was added for an additional 48 h, after which the supernatants were collected and stored at − 70°C until further processing via the multiwell plate focus assay. For each fraction and compound, two independent experiments were performed each with two replicates (n = 4). Heparin (Sigma-Aldrich) served as the positive control for viral inhibition [41].

Determining the antiviral effects of the fractions and compounds after virus cell entry
A total of 2.5 × 10 4 VERO cells were seeded in 24-well plates for 24 h, and a viral inoculum (DENV-2/NG at an MOI = 1) was then added and incubated for 2 h. Subsequently, the inoculum was removed, and the fractions (100 μg/mL) or characterized compounds (50 μg/mL) were added and incubated with the cells for an additional 48 h (post-treatment strategy) [16]. Next, the supernatants were collected and stored at − 70°C until further processing via the multiwell plate focus assay. For each fraction or compound, two independent experiments were performed each with two replicates (n = 4). Suramin served as the positive control for viral inhibition [41].

Viral inhibition assay in VERO cells
A total of 2.5 × 10 4 VERO cells were seeded in 24-well plates, and a viral inoculum was added at 24 h, followed by incubation for 2 h. Subsequently, the inoculum was removed, and the cells were treated with a single noncytotoxic concentration of each of the five fractions and compounds obtained from the fraction with the highest selectivity. At 48 h, the supernatants were removed and stored at − 70°C until further processing via the multiwell plate focus assay. Suramin served as the positive control for viral inhibition [41].

Indirect immunofluorescence
Some cell monolayers were washed with Phosphate Buffered Saline pH 7.4 (PBS, Gibco® and fixed with 3.8% paraformaldehyde (PFA, Sigma-Aldrich) in PBS at 37°C for 30 min. The cells were then treated with 50 mM NH 4 Cl for 10 min at room temperature and permeabilized with 0.3% Triton X-100, after which the nonspecific sites were blocked with 5% FBS. To detect the E protein, the monolayers were incubated sequentially with an anti-DENV-2 envelope primary mAb and an anti-mouse secondary antibody conjugated to Alexa 488. Finally, the plates were examined using a NIKON Eclipse TS100 microscope.

Molecular docking
For molecular docking, the structures of the isolated compounds and that of suramin were downloaded from the DrugBank database or were constructed from SMILES strings using CHIMERA [42]. The threedimensional structures of two viral proteins, the structural envelope protein (E, PDB: 3UZV) and the nonstructural protein 5 (NS5, PDB: 2J7U), were obtained by using the Protein Data Bank (PDB) database. Structures with a resolution equal to or less than 2.5 Å were considered. The 3D models of interest were prepared for docking using the Autodock tools package, removing both the water molecules and co-crystallized molecules not part of the target protein and adding Gasteiger charges and nonpolar hydrogens. To determine the active sites or binding sites of the protein molecules, the CASTp tool was used [43]. According to this, a grid box coordinates were defined as follows: center 2) was used to determine the best interactions between the viral proteins and compounds [44], For Autodock Vina running, a configuration file including the name of the protein with the extension *pdbqt, the ligand name with the same extension, the grid box center and the exhaustiveness, was prepared as a CONF file. Best interations were identified based on a score from 0 to − 7.0 kcal/ mol [43], with scores equal to or less than − 7.0 kcal/ mol corresponding to the best affinities. The generated interactions were analyzed using LigPlot+ v1.4.5 [45].

Statistical analysis
To determine the CC 50 and EC 50 , a regression analysis was performed with Probit Software. The SI of each molecule was determined from the relationship between the CC 50 and EC 50 , using the formula SI = CC 50 /EC 50. To compare the number of infectious viral particles released between the cells treated with each fraction or compound (for both the pre-and post-treatment strategies) and the untreated cells, Student's t-test was used.
All statistical analyses were performed using the Prism® 7.01 for Windows™ package (GraphPad Software, San Diego, CA), considering differences statistically significant at p < 0.05 in all cases. Each experimental condition was tested in triplicate in two independent experiments (n = 6) to guaranty the statistical analysis and the figures show the average of those experiments.

Antiviral effects of the ethanolic extracts
Overall, the cytotoxicity of the extracts was low (all had CC 50 values greater than 100 μg/ml). The extract with the highest CC 50 was derived from the bark of P. guajava (CC 50 = 1000 μg/mL), and the extract with the lowest CC 50 Table 2 shows these results.
Antiviral effects of the fractions obtained from the P. guajava extract Overall, the cytotoxicity of the fractions was low with CC 50 values greater than 100 μg/mL,the least toxic fraction was Pg-YP-I-22C (CC 50 = 625.7 μg/mL). Considering anti-DENV effectiveness, we found that most fractions were effective at low concentrations (EC 50 values of less than 100 μg/mL) except for Pg-YP-I-22A (EC 50 = 134.4 μg/mL). Finally, SI values were calculated for the fractions, which were then classified into the same four groups as those for the extracts. Accordingly, Pg-YP-I-22A was nonselective (SI = 1.0), Pg-YP-I-22D had low selectivity (SI = 3.2), Pg-YP-I-22E was moderately selective (SI = 6.1), and Pg-YP-I-22B and Pg-YP-I-22C were highly selective (SI = 11.7 and 35.4, respectively). Table 1 shows these results. Considering the cytotoxicity results, a nontoxic concentration (100 μg/mL) was chosen to evaluate the antiviral effects of said fractions via an inhibition test, which examined the production of infectious viral particles (Fig. 2). Only Pg-YP-I-22C and Pg-YP-I-22D significantly inhibited infection compared to the control without treatment (p < 0.05; the inhibition rates were 78.2 and 63.7%, respectively). After evaluating the SI values and virus replication inhibition percentages, Pg-YP-I-22C was chosen for further testing as its selectivity was high (SI: 35.4) and viral replication inhibition rate of 78.2%.

Identification of the compounds in the Pg-YP-I-22C fraction
Compounds 22CK001, 22CK002, 22CK003, 22CK004, and 22CK005 were isolated from the Pg-YP-I-22C fraction. Table 3 shows the results of the open column chromatography experiments. Next, the compounds were identified by one-and two-dimensional NMR spectroscopy (1D and 2D NMR) techniques and through comparisons with data reported in the literature (Fig. 3). Compound 22CK001 exhibited the following physical and spectral properties: white solid. MP 251°C. 1 C-6). This compound was identified as naringin (Fig. 3d) by comparing its 1 H-NMR and 13 C-NMR physicochemical and spectral data ( Fig. 3e and f, respectively) with those reported in the literature.
Compound 22CK003 exhibited the following physical and spectral properties: yellow needles (from methanol). MP 316°C. 1        . This compound was identified as hesperidin (Fig. 3m) by comparing its 1 H-NMR and 13 C-NMR physicochemical and spectral data ( Fig. 3n and o) with those reported in the literature.

Anti-DENV effects of the compounds obtained from the Pg-YP-I-22C fraction
The toxicity of the compounds was very low with CC 50 values greater than 400 μg/mL; catechin was the least toxic (CC 50 = 833.3 μg/mL). For antiviral effectiveness, we found that four of the five compounds were effective at concentrations lower than 100 μg/mL, with the most effective being quercetin (EC 50 = 19.2 μg/mL); hesperidin was effective only at a higher concentration (EC 50 = 225.8 μg/mL). Finally, the SI was calculated for each compound. Hesperidin was the only compound not considered selective, as it had an SI value less than 2.0 (SI: 1.8); all of the other compounds were considered highly selective, with quercetin being the most selective (SI: 34. 3). Table 1 shows these results.
Anti-DENV effects of the compounds on some steps of the DENV replication cycle Using noncytotoxic concentrations of each compound, two different experimental strategies were performed (pre-and post-treatment strategies) to evaluate their effects on processes before or after viral entry. In the pretreatment strategy, only gallic acid, quercetin, and catechin decreased infection in a statistically significant way (with viral inhibition percentages of 52.6, 50.0, and 100%, respectively). Conversely, in the posttreatment strategy, all the compounds except for hesperidin inhibited infection in a statistically significant way; the viral inhibition percentage was highest in the cultures treated with quercetin (100.0%), followed by catechin (91.8%), naringin (64.5%), and gallic acid (67.3%) (Fig. 4). These results were confirmed by decreases in viral antigen in the infected cultures treated with each compound (Fig. 5).

In silico analysis of binding between the compounds and viral proteins
The docking energy of the five flavonoids and suramin was examined using Autodock Vina. In this study, domain III of the E protein and the polymerase domain of the NS5 protein were used. All ligands showed favorable binding energies, and the interactions with the E protein were mediated by hydrogen bonds in addition to at least one hydrophilic amino acid (Ser192 and/or Lys361); only two of the five compounds, naringin (− 8.0 kcal/mol) and hesperidin (− 8.2 kcal/mol), had better scores than the theoretical threshold of − 7.0 kcal/mol with the E protein ( Table 4). The control compound suramin interacted to the E protein with a binding energy of − 7.9 kcal/mol, forming four hydrogen bonds with Val2, Lys3, Tyr106, Lys361, and to the NS5 protein with a binding energy of − 12.0 kcal/mol forming six hydrogen bonds (Fig. 6); all ligands tested except for gallic acid (− 5.3 kcal/mol) had docking energies above the theoretical threshold when interacting with the NS5 protein (Table 4), and their interactions were mediated by 1 to 7 hydrogen bonds. Among the compounds, although the hesperidin-NS5 interaction had the highest number of hydrogen bonds, the shortest distance to a hydrogen bond between the ligand and target protein appeared in the catechin-NS5 interaction (2.73 Å). The amino acids Trp477 and Gln601 participated in the interaction with gallic acid, naringin, and quercetin, while the amino acids Asn609, Asp663, and His798 were involved in the interaction with catechin and hesperidin, as well as suramin (Table 4, Fig. 6).

Discussion
At present, due to the lack of a specific antiviral drug, different strategies are being used to control symptoms in patients with dengue, such as the administration of fluids and corticosteroids and the transfusion of blood derivatives [46]. Considering the agenda of research priorities proposed by WHO as well as the biodiversity of plants in Colombia, we searched for antivirals derived from natural products that could be evaluated in preclinical/clinical studies in the medium/long term. This study first determined the selectivity of 12 ethanolic extracts derived from plants collected in the Colombian Fig. 4 Antiviral effects of the compounds derived from the Pg-YP-I-22C fraction on some steps of the DENV replication cycle. The cells were treated with the compounds for 48 h and then infected with DENV (MOI = 1). This infection strategy (pre-treatment) is represented by the dark bars. In addition, independent cell cultures were infected with DENV (MOI = 1) and then treated with the compounds (post-treatment strategy; indicated by clear bars). The error bars correspond to the SEM. The asterisks indicate cases with statistically significant differences (Student's t-test; p < 0.05) in relation to the control without treatment Caribbean coast. According to the SI values, the extracts were classified into four groups ( Table 1). The group of nonselective extracts (SI < 2.0) comprised three plants. In contrast to our results, extracts of C. citratus have been reported to inhibit DENV infection by less than 50% [47].
The differences in the results could be due to the serotype tested as well as the low specificity of the technique used in the previous report (a qualitative technique based on the cytopathic effects produced by DENV-1). An inhibition of less than 50% could indicate low selectivity, which would  be consistent with our results. Antiviral activity has not been reported for the other plants in Group A, but antioxidant activity has been reported for extracts from D. inconstans [48], and antitumor [49] and antifungal activities [50] have been documented for extracts from Ch. icaco. For the three plants in Group B (low selectivity, SI ≥ 2.0 and < 5), this study is the first to report anti-DENV activity. However, antitumor activity has been reported for C. malambo [51] and A. cumanensis [52], and lymphocyte proliferative activity has been documented for extracts from T. hirta [53]. The third group included plants whose extracts had moderate selectivity (SI ≥ 5 and < 10) and included Ch. ambrosioides, C. platanifolia, M. charantia, and M. americana. Previous reports of antibacterial and insecticidal [54], antileishmanial [55], and antimalarial activities have been described for C. ambrosioides L. [56]. Extracts of this plant have also been reported to increase nitric oxide production [57], which would favor its possible antiviral activity as increases in nitric oxide are known to lead to a decrease in DENV-2 replication [58]. For M. charantia, anti-DENV-1 activity has been reported for both extracts [47] and MAP30 protein derived from the plant [59]. Anti-DENV-2 activity has been described for two coumarins derived from M. americana [36]. This report is the first to document bioactivity for C. platanifolia. It is important to clarify that we decide to evaluate extracts from this plant, because the results from an ethnobotanical survey conducted in the city of Cartagena (Colombia) in 2009. Finally, the fourth group included the P. guajava extract, which was the only one with high selectivity (SI ≥ 10). P. guajava (which belongs to the Myrtaceae family and Myrtoideae subfamily) is a tree that grows in tropical America and tropical & subtropical regions of Asia. It is recognized mainly for its fruit, the guava [60]. The bioactivities of some parts of the P. guajava have been described for several viral infections. For example, extract from P. guajava leaves inhibits approximately 50% of human and simian rotavirus infections in vitro [61], and aqueous and methanol extracts can inhibit murine leukemia virus retrotranscriptase at percentages greater than 50% [62]. In our study, the extract of this plant proved to be highly selective (SI = 128.2) with a percentage of inhibition greater than 90% (data not shown). The inhibitory effect of leave extracts from P.guajava in culture infected with DENV has been reported previously showed a IS of 153.18 μg/mL (very similar to our results) [63]. By other hand, some studies have shown that extracts derived from P. guajava are effective against diabetes mellitus due to its antihyperglycemic activity via inhibition the enzyme α-glucosidase [64].
Accordingly, the anti-DENV effect of the P. guajava extract could be linked to the inhibition of this enzyme, as it has been described as essential for the correct folding of viral glycoproteins and for virion assembly [65] during the replication cycle. Moreover, it has been reported that other fractions derived from P. guajava can have benefits in patients infected with Dengue. For example, in culture of HepG2 cells the trombinol, a bioactive fraction of Psidium guajava, induced the thrombopoietin production, and the authors postulate that this production could be considered as an alternative treatment in patients infected with dengue [66].
In this study, five fractions were isolated from an ethanolic extract of P. guajava bark, which showed a high content of tannins, alkaloids, and flavonoids (Table 2), in accord with the phytochemical composition described by other authors for this plant [67]. Only four of the five fractions showed selectivity, with the most selective being Pg-YP-I-22C (SI = 35.4). This value was almost four times lower than that obtained with the crude ethanolicic extract (SI = 128.2), which could indicate that the greater effectiveness of the extract may be due to synergy.
From the most selective fraction (Pg-YP-I-22C), five compounds were isolated and identified (Table 3 and Fig. 3). The first of these, gallic acid, is a polyphenolic compound [68] with diverse biological activity. Our results showed that gallic acid significantly inhibited viral activity via both the pre-and post-treatment strategies (percentages of inhibition greater than 50%). Previously, a cocktail (containing several compounds including gallic acid) derived from plants in the Phyllanthaceae family was shown to have an antiviral effect against DENV [69], unfortunately our results cannot be compared with those previously reported, as the effect of one compound alone is not comparable to that produced by the synergy of several compounds. However, recently has been reported that the isobutyl gallate (a gallic acid derivative) is an antiviral against DENV-2, with a SI = 25.6 in Huh 7 cells quite similar to our results in VERO cells (SI = 21.1) [70]. The second compound, naringin, is also a flavonoid with diverse biological activity [71]. In our study, naringin significantly inhibited DENV-2 infection only it was added to cells after viral inoculation (post-treatment). The anti-DENV activity of naringin has been previously demonstrated [72], but such inhibition occurred only when viral inoculation of the cells was performed in the presence of the compound (a method we did not use, thus making the results incomparable). This previous report did not investigate viral inhibition when the treatment was performed after inoculation, in contrast with our results. However, it is important to note that the concentrations we used were higher than those previously reported without being cytotoxic and that our SI (SI = 13.5) was ten times higher than that previously reported (SI = 1.3). The third compound, quercetin, is another flavonoid with diverse biological activity [73], including anti-DENV activity [72]. Although our results agree with those previously reported, we emphasize that the previously demonstrated selectivity (SI = 7.5) is much lower than that reported by us (SI = 34.3), which may be because we tested a higher concentration than that used previously. Quercetin has been postulated to directly inhibit the viral NS3 protein (a protein with multiple roles in DENV replication) [74], and it could interrupt virus entry by inhibiting fusion [75], making this compound a very promising antiviral. Catechin, the fourth compound isolated, is a polyphenol that is mainly derived from green tea [76]. Of the five compounds identified in this study, catechin is the one that induced the best viral inhibition when added before (100% inhibition) or after inoculation (91.8%). Recent molecular docking studies have shown that catechin has a high binding affinity for the NS4B protein of DENV [77], which is an important protein in the formation of the viral replication complex (together with NS4A) in the endoplasmic reticulum of host cells [78]. Finally, the fifth compound, hesperidin, is a flavonoid [79] of which no antiviral effect has been reported, agreeing with our results. It was the only nonselective compound (SI = 1.8) and did not inhibit infection in any of the experimental strategies used.
In addition to the in vitro studies, we used computational tools to explore the possible mechanisms involved in viral inhibition. For this, the E and NS5 proteins were selected as targets for the compounds tested, as in vitro inhibition was found for pre-and post-entry steps and their three-dimensional structures were available for in silico tests. Using computational methods, new antiviral molecules targeting the NS3 protease have already been reported that, when evaluated in vitro, inhibited DENV-2 infection up to 1 log PFU/mL [80], demonstrating the power of these methods in facilitating the search for new antiviral molecules.
In our study, for the E protein, the structure used corresponded to domain III of DENV-2; however, the crystal structure probably corresponds to the mature protein, as it was co-crystallized with the variable portion of monoclonal antibody 4E11 [81]. Nonetheless, based on the in vitro experiments, the molecules tested might influence the immature protein, possibly affecting the interaction by molecular docking. For the NS5 protein, even though the crystal structure corresponds to the catalytic domain [82], the binding pocket was defined using computational tools; thus, the docking energy values could vary. However, the negative docking energy values suggest that all the interactions were favorable [83]. Notably, the interaction of suramin with both proteins was favorable, but the best docking energy was found for its interaction with NS5 (− 12.0 kcal/mol), which was mediated by six hydrogen bonds (Fig. 6). This result is consistent with the data from the in vitro experiments in which better inhibition was observed for this compound via the post-treatment strategy (Fig. 4), suggesting that this compound acts on viral polymerase.
The docking energy values depend not only on the binding site but also on the virus type, target protein, and ligand source. A consensus model of the NS5 protein from four DENV serotypes was used previously against natural compounds obtained from the PubChem database22 and the SuperNatural II database2, yielding binding energies of < − 10.5 kcal/mol [84]; conversely, the compounds tested herein were obtained by conventional bioprospecting methods. The hydrogen bonds and hydrophobic interactions between the target protein and ligands determine the molecular docking score [85], and the distance between the atoms that are part of the bond is of great biological importance, especially if they range between 2.5 and 3.5 Å, as those with small distances are more relevant [83]. Such is the case of the hesperidin-E interaction, which has 11 theoretical hydrogen bonds compared to the four theoretical hydrogen bonds formed in the catechin-E interaction (Table 4).However, the shortest distance between the atoms forming a hydrogen bond in the proteinligand pairs was obtained in the catechin-E interaction (2.70 Å). The distance between atoms was also the lowest (2.73 Å) in the catechin-NS5 interaction compared to those of the other interactions evaluated. Similar studies have reported that catechin can interact with other viral proteins such as NS4B from DENV-2 with negative docking scores (− 4.06 kcal/mol), which suggests that this interaction would be viable [77]. Other factors that can affect the docking score are ligand size and the number of atoms forming bonds. Consequently, smaller ligands such as gallic acid, quercetin, and catechin have lower docking scores and lower numbers of hydrophobic interactions compared to those of naringin and hesperidin (Table 4).
In spite of the importance of ENV and NS5 proteins for viral replication, well-conserved non-structural proteins such as NS1, NS3/NS2B have been also evaluated against phytochemicals reported in literature, and the evaluated interactions were favorable with scores ranging from − 6.71 to − 32.24 Kcal/mol according to MOE Dock tool [86] Similarly, NS3 was evaluated against the compounds derived from the medicinal plant Vetiveria zizanioides, and one compound showed favorable interactions according to the scores obtained with Surflex-dock, mediated by Hydrogen bonds with 2.72 Å being the lowest distance found mediating this interactions [87].
Although bioprospecting has been replaced by more targeted tools, it is still useful for identifying antiviral compounds, as we have shown in this study. Additionally, combining in vitro and in silico tests allowed us not only to identify promising antivirals but also to suggest their possible mechanisms of action.

Conclusions
Our results showed that four compounds derived from P. guajava highly selectively inhibited DENV-2 replication. Catechin is the most promising compound with viral inhibition percentages of greater than 90% in the two different experimental strategies. Studies are in progress that will allow us to elucidate the antiviral mechanisms of these compounds.