Aqueous solubility.

Being one of the chief physicochemical properties to be optimized in drug discovery, aqueous solubility has been considered as an important factor in drug absorption and distribution. And one drug with adequate aqueous solubility is considered exempt from bioavailability problems [18]. Here, for filtering compounds with suitable aqueous solubilities from herbs, we calculated the logS of each molecule using the ALOGPS 2.1 program [19]. This model was developed with 1291 molecules using ANN methodology and electrotopological state descriptors, resulting in an optimal predictive model with determination coefficient (R²) = 0.91 and root mean squared error (RMS) = 0.62. Considering the fact that most drugs reveal a compromise between the polarities required for reasonable aqueous solubility and the hydrophobicities needed for satisfactory membrane passage [20], we set the threshold of logS in the range of −5 to −1.

Lipophilicity.

Molecular lipophilicity, another basic physicochemical property, plays an essential role in determining ADME properties and the overall suitability of drug candidates, hence controlling molecular lipophilic property within an optimal range can improve compound quality and the likelihood of therapeutic success [21]. In this text, molecular lipophilicity (expressed as a value of logP) was also predicted by ALOGPS 2.1 program, which developed the predictive model of lipophilicity on the basis of 64 neural networks, 12,908 compounds with experimental logP values from PHYSPROP database and 75 input parameters, providing an optimal model with RMS = 0.49 and standard mean error = 0.26 [22]. On the basis of Lipinski's rule of five, the compounds with logP less than 5 were selected for further analysis.

Drug-likeness.

For the reason that medicinal properties of herbs depend on the presence of active constituents with drug-likeness features, in this study, we employed a robust self-constructed model preDruglike as described in our previous work [23] to calculate the drug-likeness index of each compound in herbs. This model is constructed based on the molecular descriptors and Tanimoto coefficient (as displayed in Eq.2).(2)where A shows the molecular properties of herbal ingredients, and B displays the average molecular properties of molecules in DrugBank database (http://www.drugbank.ca/) based on Dragon soft descriptors. Considering the fact that the average DL index for all 6511 molecules in DrugBank is 0.18, the molecule with suitable drug-likeness index (DL≥0.18) was chosen as candidate molecule for further research.

Oral bioavailability.

Being an essential parameter in drug screening cascades, oral bioavailability was employed to determine the fraction of oral dose of compounds reaching systemic circulation in the TCM treatment. Here, a reliable in silico screening model OBioavail 1.1 constructed in our previous work [24] was engaged in OB value calculation of these herbal constituents. This model was constructed based on 805 structurally diverse drugs and drug-like molecules. Multiple linear regression, partial least square and support vector machine methods were employed during this model building, ending up with determination coefficient (R²) = 0.80 and standard error of estimate (SEE) = 0.31 for test sets. Finally, the threshold of OB value is set to 30% by careful consideration of the following rules: 1) extracting information from the studied herbs should be as much as possible using the least number of molecules [13]; 2) the acquired model can be reasonably explicated by the existing pharmacological data [14].

Caco-2 permeability.

For an orally administered drug, the majority of drug absorption occurs in the small intestine where the presence of villi and microvilli greatly increases the surface available for absorption [25]. A number of in silico drug absorption models using in vitro Caco-2 permeability have been widely established and used in drug discovery and development processes [26]. Here, we employed a robust in silico Caco-2 permeability prediction model preCaco2 [27] which was constructed by 100 drug molecules with satisfactory statistical results (R²>0.8) to predict the drug absorption. Finally, we set the threshold of Caco-2 permeability to −0.4, for the reason that compound with Caco-2 value less than −0.4 is not permeable.

BBB penetration.

Since blood-brain barrier is a protective fence between central nervous system (CNS) and systemic circulation to maintain the normal homeostasis of the CNS [28], predicting BBB permeability is regarded as an indispensible parameter in screening candidate compounds for stroke management. In this section, an updated and reliable BBB model preBBB which was constructed in our previous work [29] was employed to examine whether the constituents of herbal medicines pass through the BBB. The dataset of this model was composed of 190 related but chemically diverse compounds which are either penetrating or non-penetrating cross the BBB [23]. The statistical model was constructed by partial least squares discriminant analysis with two significant latent variables. In this model, compounds possessing BBB values greater than −0.3 were considered as penetrating, thus the threshold of BBB is set to −0.3.

Uncaria rhynchophylla.

In medicinal herb U. rhynchophylla, 41 out of 92 compounds are found meeting the screening criteria. Among them, plenty of indole alkaloids with satisfactory ADME properties have already been demonstrated exerting significant protective effects against several diseases. For example, the major constituents of this herb rhynchophylline (logP = 2.85, logS = −3.66, DL = 0.57, Caco-2 = 0.70, OB = 47.86% and BBB = 0.38) and isorhynchophylline (logP = 2.47, logS = −3.36, DL = 0.52, Caco-2 = 0.29, OB = 49.31% and BBB = −0.21) have been reported showing neuroprotective, vasodilatory and antiarrhythmia effects [41], [42], thus becoming the candidate compounds for treating cardiovascular and central nervous system diseases. Furthermore, isorhynchophylline also exhibits inhibitory effects on platelet aggregation and thrombosis, which mechanism may be at least partly due to the increase of intraplatelet cAMP generation [42], [43]. It is noteworthy that two bioactive compounds kaempferol (logP = 1.23, logS = −2.47, DL = 0.24, Caco-2 = 0.28, OB = 42.30% and BBB = 0.02) and quercetin (logP = 1.07, logS = −2.42, DL = 0.28, Caco-2 = 0.03, OB = 42.24% and BBB = 0.00), being the common flavones existed in several herbs like U. rhynchophylla and G. biloba, also show satisfactory pharmacokinetic properties.

As we know, unlike allopathic western medicine, herbal medicine characterized as holistic treatment emphasizes on maintaining the integrity of human body. Factually, medicinal herb U. rhynchophylla not only acts on the central nervous system to produce anti-inflammation, anti-apoptosis and free radical scavenging activities [44], but also exhibits antiplatelet and anti-hypertensive effects on the cardiovascular system [45]. These phenomena prove that, as a typical holistic therapy, herb U. rhynchophylla, besides providing benefits in stroke treatment, shows preventive effects in primary and secondary stroke prevention. Therefore, compounds which have suitable pharmacokinetic properties but low BBB values are also considered as potential active compounds, such as L-epicatechin (logP = 1.02, logS = −2.65, DL = 0.24, Caco-2 = 0.01, OB = 53.89% and BBB = −0.65) and angustoline (logP = 1.84, logS = −3.28, DL = 0.77, Caco-2 = 0.41, OB = 33.15% and BBB = −0.31).

Scutellaria baicalensis.

For herb S. baicalensis, 56 out of 111 constituents pass through the filtering criteria, and the majority of them (46 out of 56 active compounds) are flavonoids, such as oroxylin A (logP = 3.01, logS = −3.34, DL = 0.44, Caco-2 = 0.77, OB = 45.59% and BBB = 0.13) and acacetin (logP = 2.95, logS = −3.51, DL = 0.35, Caco-2 = 0.62, OB = 34.93% and BBB = 0.00). Surprisingly, most flavonoids in this herb have been proven to possess various biological activities. For instance, oroxylin A shows memory ameliorative activity in memory impaired mice [46], and also has neuroprotective effect against ischemic/reperfusion-induced brain damage [47]; acacetin exhibits anti-neuroinflammation effect through regulating the response to LPS stimuli in vitro and in vivo, and is also considered as a potential therapeutic agent for brain diseases [48].

Besides these, those constituents which have high contents in herb but low OB indexes such as baicalin (OB = 13.16%), baicalein (OB = 23.77%) and wogonin (OB = 20.71%) should attract our more attention. For example, baicalin obtains a poor OB index, however, the content of this flavone in S. baicalensis is extremely high (8.12% of dry root mass) [48], which significantly increase its absolute OB value. Actually, baicalin has been documented exhibiting powerful pharmacological activities, including protective effect against cerebrovascular dysfunction and inhibitory effect on brain inflammatory response [49]. Therefore, these bioactive ingredients with high contents in herb are also added into active ingredients database, although they have low OB values.

Salvia miltiorrhiza.

After ADME prescreening, 35 constituents from S. miltiorrhiza have desirable physicochemical and physiological properties, including many documented bioactive compounds such as cryptotanshinone (logP = 4.32, logS = −4.48, DL = 0.40, Caco-2 = 0.95, OB = 57.44% and BBB = 0.51) and tanshinone IIb (logP = 3.31, logS = −4.01, DL = 0.45, Caco-2 = 0.43, OB = 70.19% and BBB = −0.30) which have been reported to show neuroprotective [50], antidiabete [51], anti-atherosclerosis [52] and anti-inflammatory [53] effects. Two bioactive compounds tanshinone IIa and tanshinone I as the major constituents of S. miltiorrhiza are also put into the active ingredient database due to their in vitro and in vivo biological activities [54]. One exception is salvianolic acid B which obtains the low OB (3.01%), caco-2 (−1.68) and BBB (−2.52), however, several in vitro studies indicate that salvianolic acid B displays significant pharmacological activities: protection of local cerebral ischemia-reperfusion injury [55], inhibition of platelet aggregation [56] and low density lipoprotein oxidation [57], as well as improving regional cerebral blood flow. In fact, several investigations have proposed that the bioactive effects of salvianolic acid B may not only be due to itself but also its metabolites, like isoferulic acid (logP = 1.56, logS = −2.35, DL = 0.06, Caco-2 = 0.50, OB = 67.69% and BBB = 0.08) and salvianic acid A (logP = 0.77, logS = −1.47, DL = 0.06, Caco-2 = −0.27, OB = 78.22%, BBB = −0.62) [58], [13]. Therefore, these metabolites are also added into active ingredients database for further research.

C-T network: elucidating the combining and additive strategies for herbal medications in stroke therapy and prophylaxis.

As displayed in Figure 5, C-T network is generated based on 641 associations between 168 herbal ingredients (magenta circle nodes) and their corresponding 94 targets. As outlined above, we distinguish the potential targets into stroke therapy (limegreen circle nodes) and prophylaxis (limegreen diamond ones). Network analysis results show that the average number of targets per compound is 3.9. Nearly two thirds compounds (105 out of 168 compounds) are linked with more than one target, showing the multi-target properties of herbal ingredients. Among them, compound quercetin possesses the highest number of target interactions (degree = 42), following by apigenin with 26 targets and luteolin having 25 linked targets.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Compound-target network of anti-stroke herbs.

Node size is proportional to its degree.

https://doi.org/10.1371/journal.pone.0102506.g005

As mentioned above, several pathways that lead to cell death are triggered by cerebral ischemia; therefore effective neuroprotective strategy might require a combination of drugs which act on distinct pathways during the progression of ischemic damage. Fortunately, by further analysis of the C-T net, we conclude that, in one herb, several compounds may bind to different points of same signaling pathway or different targets of different signaling cascades to exhibit the synergistic therapeutic actions. As an illustration, herb Ginkgo biloba (as displayed in Figure 6) was specified in detail to interpret the combination principle. During ischemia and reperfusion injury, overactivation of PPAR1 can induce translocation of AIF (apoptosis-inducing factor) from mitochondria to nucleus, which results in subsequent cell death [67]. Fortunately, compounds quercetin and myricetin of G. biloba which are identified as significant inhibitors of PPAR1 [68] collectively mediate the caspase-independent pathway, and have the potential to exhibit the additive anti-apoptosis effects. Moreover, compounds ginkgolide B and bilobalide of this herb can block neuronal apoptosis by attenuating the activation of caspase-3 [69]; meanwhile myricetin can directly bind to the active site of caspase-3 to inhibit its activity [70]. All of these document that this herb is possible to regulate the caspase-independent pathway to display the synergistic inhibition of neuronal cell apoptosis. Therefore, we speculate that herb G. biloba might provide an effective therapeutic approach for the treatment of cerebral ischemia through synergistic inhibition of both the caspase-dependent and independent pathways.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 6. Illustrating the synergistic actions of anti-apoptosis of Ginkgo biloba.

https://doi.org/10.1371/journal.pone.0102506.g006

Further analysis of the C-T net shows that, besides those combination therapies, several active ingredients in one herb can act on one common target, which might exhibit additive effects for improving the stroke outcome. For instance, after ischemia develops, NOS2 which produces NO contributes to the evolution of the cerebral injury [71], because of the fact that NOS2 is expressed in the setting of the inflammation response after cerebral ischemia. Then, inhibition of NOS2 expression or activity has potential to develop into an attractive therapeutic strategy for stroke. Fortunately, 39 herbal ingredients (as shown in Figure 5) like kaempferol, quercetin, ginkgolide B, ginsenoside C-K and oroxylin A all have interactions with NOS2, and might contribute to the anti-stroke effect of herbal medicines. Just for herb S. baicalensis, 12 constituents such as oroxylin A, acacetin and eugenol work together on protein NOS2, and might provide additive effects to reduce the infarct volume and ischemia brain damage.

Owing to the pathophysiology of stroke involved in multiple mechanisms, the use of combinations for improving the stroke outcome is especially rational. Facts have proven that therapies that eliminate the clot, restore blood flow, inhibit excitatory neurotransmission, reduce inflammatory response following ischemia/reperfusion, or scavenge free radicals all have promising therapeutic potentials in animal models of stroke [72], [73]. Luckily, in the C-T network of anti-stroke herbs (Figure 5), several high degrees of correlations of targets for anti-inflammation (like NOS2, ALOX5 and PPARG), anti-oxidant (including XDH and HMOX1) and anti-apoptosis (such as BAX, BCL2 and caspse-3), neuroprotection (for instance: AKT1, MMP9 and MMP2) are regulated by herbal ingredients, which might display a combination of multiple mechanisms for effective long-term treatment for stroke patients.

Furthermore, as we know, the incidence of stroke can be reduced with appropriate preventive measures like blood pressure lowering, as well as lipid-lowering, anti-diabetic, anti-thrombotic and antiplatelet therapies. In the C-T network of Figure 5, besides those potential therapies for stroke treatment, series of target proteins of herbal constituents are also engaged in diabetes mellitus (such as PPARG and NR3C1), atherosclerosis (for instance: PLA2G2A, NR1H3 and F10), hypertension (including ADRB1, NOS3 and ACE), dyslipidemia (like NR1H2, PPARD and HMGCR), which may be implicated in the primary and secondary stroke preventions.

In summary, as displayed in the C-T net, botanical medicines are engaged in multiple mechanisms of stroke to synergistically improve outcome, which are roughly classified into preventing inflammatory response, suppressing apoptosis, scavenging free radicals, inhibition of platelet aggregation, as well as modifying diabetes mellitus, hypertension, dyslipidemia and other stroke risk factors. And several ingredients in one herb might act on one the same target to display an additive effect, or bind at the different points of the same signaling pathway or at different targets of several pathways, and then have the potential to exhibit the synergistic therapeutic actions for stroke therapy and prophylaxis.

T-P network: holistic mechanisms of anti-stroke medicinal herbs.

For better elaborating the major pathways involved in herbal medicines for stroke therapy, we extract the canonical pathways that are highly associated with stroke from KEGG database (http://www.genome.jp/kegg/), resulting in 30 canonical pathways including PI3K-Akt signaling pathway, TNF signaling pathway, neuroactive ligand-receptor interaction, calcium signaling pathway and metabolic pathways. For example, being a central mediator in signal transduction pathways modulating cell growth, metabolism and survival, the PI3K-Akt signaling pathway can mediate the neuroprotective activity of vascular endothelial growth factor and induce BBB permeability after focal cerebral ischemia [74]; MAPK signaling pathway transduces a large number of external signals, resulting in a large-scale cellular responses which contain cell proliferation, differentiation, inflammation and apoptosis, and now it is becoming obvious that this pathway plays a significant role in mediating cell survival following brain ischemia.

Subsequently, all target proteins from target fishing and validation attempt to be mapped onto these 30 pathways, generating a bipartite target-pathway network graph as displayed in Figure 7. Results show that after discarding 9 target proteins without participating in these pathways, this T-P network consists of 119 nodes (85 targets and 30 pathways) and 280 edges. And these pathways are linked with target proteins intensively, such as the TNF signaling pathway which exhibits the largest number of interactions with targets (degree = 17), neuroactive ligand-receptor interaction (degree = 16), PI3K-Akt signaling pathway (degree = 16) and HIF-1 signaling pathway (degree = 14). Actually, plenty of compounds in medicinal herbs are involved in these pathways, which maybe provide basis for stroke treatment and prevention strategies as well. For example, 62 herbal ingredients like ephedrine, corynantheine, ginkgolide A and pseudoephedrine are involved in mediating the major components of neuroactive ligand-receptor interaction pathway including ADRA2A, DRD2 and GABRA2, and might provide neuroprotective strategies to cerebral ischemia reperfusion injury. Analogously, the activation of PI3K-Akt signaling pathway has been proved essential for preventing the neuronal apoptosis and protecting the brain from cerebral ischemia/reperfusion injury [75]. And herbal ingredients like salvianolic acid B [76], cryptotanshinone [50] and ferulic acid [77] have been documented to regulate the PI3K-Akt signaling pathway in cerebral ischemia, thereby exhibiting protective effects against brain injury. Additionally, it is worth noting that several other pathways such as PPAR signaling pathway, arachidonic acid metabolism, insulin signaling pathway and TGF-beta signaling pathway can also be observed in this T-P network, and these pathways are closely related with glucose homeostasis, lipid profile, platelet coagulation and blood pressure. Therefore, we speculate that herbal medicines probably mediate these pathways to exhibit the anti-diabetes, antihypertensive, anti-platelet and anti-hyperlipidemia properties, and thereby might provide a combining system for stroke prevention.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 7. Target-pathway network of anti-stroke herbs where magenta nodes represent the targets and limegreen nodes signify pathways.

Node size is proportional to its degree.

https://doi.org/10.1371/journal.pone.0102506.g007

Consequently, we deduce that herbal ingredients mainly target on proteins such as PTGS2, NOS, GSK3B, F2 and CASP3, and thereby have the potential to regulate the pathways like TNF signaling pathway, neuroactive ligand-receptor interaction, PI3K-Akt signaling pathway and HIF-1 signaling pathway to show the anti-apoptosis, anti-oxidant, anti-inflammatory, as well as other neuroprotective effects. Besides these, constituents of these ant-stroke herbs might also disturb PPAR signaling pathway, arachidonic acid metabolism, insulin signaling pathway and other pathways via regulating PPARG, MAPK1, PTGS1 and so forth, and then display anti-diabetes, antihypertensive, anti-platelet and anti-hyperlipidemia properties against recurrent stroke.