SEARCH

SEARCH BY CITATION

Keywords:

  • drug design;
  • multitarget drug;
  • quantitative composition-activity relationship;
  • traditional Chinese medicine

Abstract

  1. Top of page
  2. Abstract
  3. Methods
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

Traditional Chinese Medicine has become an important resource for searching the effective drug combinations in multicomponent drug designs. In this article, we investigate the methodology on how to efficiently optimize the combination of several active components from traditional Chinese formula. A new method based upon lattice experimental design and multivariate regression was applied to model the quantitative composition-activity relationship (QCAR) in this study. As a result, multi-objective optimization was achieved by Derringer function using extensive search algorithm. This newly proposed QCAR-based strategy for multicomponent drug design was then successfully applied on search optimal combination of three components from Chinese medicinal formula Shenmai. The result validated the effectiveness of the presented method for multicomponent drug design.

Tremendous progress has been made in the pharmaceutical industry and many therapeutic agents have been developed to target specific proteins over the past two decades. Nonetheless, single agent therapy has been proven ineffective to treat cancer, diabetes and other complex diseases because of the diversities of cellular functions and interactions among the pathways, which may limit the performance of single-target-drug paradigm (1). Meanwhile, some researchers found that a combination of multiple compounds, in many cases, can achieve more effectiveness than using a single compound (2,3). Thus, the production of effective combinatorial mixtures to manipulate multiple disease targets has gained increasing attentions in the research community of drug discovery and development.

A few different design strategies for multiple components drug have been reported in literature. For example, high-throughput screening method has been used for identifying effective combinations of clinically used compounds (4). Systematic screening of about 120 000 different two-component combinations only resulted in dozens of confirmed pairwise combinations. It attempted to use computational method to reduce the size of sample sets as well as the labor- and time-consuming biologic assays. Wong et al. (5) proposed a stochastic search algorithm to choose only tens of screens out of a large number of possible combinations. Alternatively, others suggested that drug regimens that contain several active components, such as traditional Chinese medicine (TCM) and other herbal medicine, can be important natural sources of drug candidates or multicomponent drugs (6,7). In fact, those natural mixtures from herbs and herbal extracts have been widely applied for the treatment of various diseases for hundreds of years until the monumental discoveries of the sulfonamides and penicillin. It looks like it is the right time to design multicomponent drugs from natural products by integrated use of past medical experiences and modern techniques.

Developing an optimum composition of several active components from TCM for the maximum therapeutic efficacy with minimum side-effects is a critical step in the discovery of multicomponent drug from natural sources, which generally involves two steps. In the first step, the active components of TCM are screened and identified. The primary goal in this step is to validate the efficacy of multiple ingredients and narrow the testing parametric space in the subsequent mixture design. Once the active components critical to the therapeutical effects are found, the second step of combination optimization is to find the optimum proportion of each component for single or multiple goals. If only two active components are combined, the optimal proportion can be easily screened by a set of experiments. However, when a combination consisted of three or more components needed to be optimized, the large parametric space imposes a major challenge to experimental design and data analysis. The application of statistical experimental design methods and mathematical models may greatly contribute to the effective searches for the potent drug combinations.

Computational approaches, which generate structure- or ligand-based models to predict bioactivity, have been found to be valuable in the optimization and development of potent drug candidates (8). Many studies have shown that mathematical and computational approaches, such as graph/diagram analysis (9), pharmacophore modeling (10), structural bioinformatics (11), molecular docking and packing (12), Mote Carlo simulated annealing approach (13), bio-macromolecular internal collective motion simulation (14), quantitative structure-activity relationship (QSAR) (15), protein subcellular location prediction (16,17), identification of enzymes, proteins, proteases and their types (18,19), and cellular automaton analysis (20), can timely provide very useful information and insights into both basic research and drug design and hence are widely welcomed by science community. But the structural information of individual components is rarely related to the activity of multicomponent mixture. The key optimization task for multicomponent drug design that has to be solved is to determine the proper proportion of each component in the optimal combination. In our earlier study, we had proposed computational approaches to modeling the quantitative composition-activity relationship (QCAR) (21,22), and screening optimal proportion of two active components out of six components from a Chinese medicinal formula (23). The approach developed a relational model between biologic potency of a combination and its chemical composition (i.e. proportion and dose of individual component). However, such model still needs to be improved because of the large number of tested samples used as training set. The present study attempts to develop a new computational approach for more efficiently conducting multicomponent drug design.

The purpose of this article is to experimentally implement a QCAR-based scheme for multicomponent drug design from TCM. A simple lattice design is applied to obtain a small set of combinations. Multi-objective optimization of three components combination from a traditional Chinese formula was achieved. The approach proposed in this study effectively identified potential drug combinations toward the goal of integrative therapy.

Methods

  1. Top of page
  2. Abstract
  3. Methods
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

Principles of QCAR-based multicomponent drug design

In modern pharmaceutical industry, computer-aided drug design methods, such as QSAR has greatly accelerated the pace of drug discovery in recent decades (24–26). The principle of QSAR modeling is that the variation of biologic activity within a group of compounds is correlated with their 2D/3D structural features as well as binding energy, topological index, and force filed (27,28). Figure 1A shows the QSAR design paradigm to find out optimal compound (drug candidate) from a set of chemicals with similar structural features. The computational modeling plays a pivotal role in bridging chemical property descriptors and biologic functions. The discovery of multicomponent drug can be achieved in the same manner that parallels current QSAR method. In Figure 1B, an illustration about designing mixture of three components was exhibited. Suppose three different components have distinct colored property shown as red, yellow, and blue respectively. According to the Young–Helmholtz trichromatic theory, the combination of three pigment colors can produce all visible colors (29). We can see that the combination of blue and yellow equals green, while red plus blue produces purple. Therefore, all possible combinations construct a large parametric space, in which each point represents one assembled mode. The task for the selection of optimal combination depends on what kind of color is most preferred. When the color of points was instead of multiple bioactivities, the optimal combination (here we called ‘multicomponent drug candidate’) can be designed for the purpose of obtaining best biologic activities and minimal side-effects.

image

Figure 1.  Illustration of different strategies for drug design. (A) The commonly used QSAR approach. (B) The QCAR approach for multicomponent drug design.

Download figure to PowerPoint

Algebraically, the process of multicomponent drug design can be interpreted as follows. Suppose n kinds of different components were used to combine multicomponent mixtures and the dose of each component is Ci(i = 1,2,...n). If the proportion of each component in a certain combination R is represented by a vector inline image, then composition of R can be represented by a vector inline image, where inline image is calculated by

  • image(1)

When m kinds of biologic activities (including therapeutical efficacy and side-effects) for combination R are assayed, the biologic properties of R can be denoted by column vector inline image. Thus, {XR, YR} represents a combination mode in the composition-activity space. For another mixture R ′, its composition and bioactivities can be described as {XR ′, YR ′}. If we collect enough samples with known chemical and biologic data used as training set, the relationship between chemical composition and biologic activities of the multicomponent combination can be quantitatively modeled by the following mathematic function:

  • image(2)

Therefore, the task for multicomponent drug design can be transformed into two steps. The first step is developing appropriate models to quantitatively interpret the relationship between combinations and their activities, which is a pattern recognition or function approximation problem in parallel with QSAR study. The second step is using the proposed model to simulate a large-scale composition-activity space and search potential combination in silico.

Experimental design for QCAR-based multicomponent drug design

Various statistical approaches as well as artificial intelligence techniques can be employed to establish linear or non-linear QCAR models. But the performance of models not only depends on the capability of computational algorithms, but also is vastly affected by the reliability and quality of training set. Appropriate experimental design might contribute to construct an ideal training set.

Numerous experimental design methods have been applied in formulation design and molecular selection (30,31). An experiment design E can also be used to arrange appropriate combinations of several components. Algebraically, E is a m × n matrix, where m is the number of tested combinations, Ejk(j ∈ [1, m], k∈ [1, n]) equals the proportion of component Ck in combination j, which is in the range of [0,1]. Thus, the chemical composition information of m combinations can be calculated by X× n = E × X, where X = [X1, X2,...,Xn] is the dose of each component. After bioactivities of combinations were assayed, the data set containing m groups of {X, Y} can be used for modeling the quantitative composition-activity relationship of those combinations.

In the current study, the optimization of three components mixture was performed. A simple lattice design was employed to arrange the experiments. The details on simple lattice design have been given in many refs (32). Briefly, a {k, j} simplex lattice design consists of all possible combinations of 0, 1/j, 2/j,...j/j or a total of = (k+j−1)!/[(k−1)!j!] experiments where there are k factors. A {3, 3} simplex lattice design used in present study includes three single factor experiments, six experiments where one factor is at 2/3 and the other at 1/3, and one experiment where all factors are at 1/3. The combination of a three-component system is represented by an equilateral triangle in two-dimensional space in which red spots are tested samples (Figure 2).

image

Figure 2.  Three factor simplex lattice design for the optimization of Shenmai Formula.

Download figure to PowerPoint

The optimization task in optimal combination design

After quantitative chemical composition-activity relationship is modeled based on the proposed datasets, appropriate optimization strategy should be established to search potential optimal combinations. As the number of combinations is essentially limitless, a set of criterions for determining the most promising combinations and prioritizing their evaluation are crucial. Unlike QSAR study aiming at the efficacy on single targets, QCAR study needs to consider several kinds of biologic activities, e.g., therapeutic efficacy and side-effects. Therefore, the optimization can be performed in two ways. Response surfaces can be simulated for each biologic activity, and these surfaces can be analyzed simultaneously by multi-objects optimization methods. Various approaches such as evolutionary and genetic algorithms have been used to solve such multi-objective optimization problems (33). The second way is to establish a model for a single composite function that takes into account all responses to obtain an integrated response surface. Then multicomponent drug candidate can be found by extensive or stochastic search on the integrated response surface.

In current study, we integrated multiple activities by a desirability function D, which was proposed by Derringer and Suich in 1980 (34) and recently obtained increasingly popular interests for treating multiple responses (35). In short, after individual response surfaces are determined for each biologic activity, predicted values obtained from each response surface are transformed to a dimensionless desirability scale di. The scale of the desirability function ranges between d = 0, for a completely undesirable response, and d = 1 for a fully desired response, above which further improvements would have no importance. For the therapeutic efficacy, in which one wants to maximize the response, eqn (3) is applied in which Y+ and Y represents the maximal and minimal preferred efficacy respectively. When minimization of a side-effect is required, the opposite of these if/then rules are applied.

  • image(3)

After transforming each criterion, the overall efficacy D is calculated by the geometric mean: D = (d1 × d2 × ...dm)1/m. The combination with highest value of D is regarded as globally optimal combination for the therapeutic efficacy.

Data sets

In this study, Shenmai formula, one of the most commonly used traditional Chinese medicines for the treatment of coronary atherosclerotic cardiopathy and viral myocarditis (36), is used as an example to illustrate the proposed method. The major active components in Shenmai formula are panaxadiol (PD) and panaxatriol (PT) type saponins from Red Ginseng and ophioponins (OP) from Radix Ophiopogonis (37). Many studies have proven that PD and PT exhibit distinct pharmacological effects, although they have similar chemical structures (38).

The simplex lattice design was adopted to arrange combinations of three components with variant proportion. Seven combinations as well as three individual components were used for pharmacological assays. The infarction area of heart and serum malondialdehyde (MDA) level in acute ischemia rat were used to evaluate the therapeutic effect of each component or component mixture. Chemical composition and bioactivity data of the sample sets were shown in Table 1. The following is a brief description of the pharmacological experiment.

Table 1.   Combinations in simplex lattice design and their cardioprotective activity
Group (n = 8)Proportion of componentsInfarct rate of heart (%)MDA (nmol/mL)
PDPTOP
  1. Compared with MI group, *p < 0.05, **p < 0.01.

Model26.72 ± 3.9910.00 ± 1.91
C11.0000.0000.00023.24 ± 3.148.39 ± 1.12
C20.0001.0000.00021.65 ± 0.987.22 ± 0.75
C30.0000.0001.00022.40 ± 3.647.75 ± 1.19
C40.6670.3330.00020.21 ± 5.618.24 ± 1.19
C50.6670.0000.33321.48 ± 5.986.13 ± 1.60**
C60.3330.0000.66720.75 ± 2.266.34 ± 0.85**
C70.3330.6670.00020.45 ± 5.326.50 ± 2.20
C80.0000.3330.66714.21 ± 5.76*9.29 ± 0.75
C90.0000.6670.33314.32 ± 2.72**6.32 ± 1.30*
C100.3330.3330.33312.56 ± 4.07**6.37 ± 2.40

Male Sprague–Dawley rats (∼200 g) were used to perform myocardial infarction, which is produced by occlusion of the left anterior descending (LAD) coronary artery according to the method of Yamaguchi (39). Rats were randomly assigned to control group, myocardial infarction group, and several treated groups. Each group contains at least eight animals. The mixture was daily given by oral administration for 3 days after LAD occlusion. The dose compares well with the dose in humans and was confirmed by our preliminary study. The infarction rate as a percent ratio of the left ventricular mass was stained by triphenyltetrazolium chloride. The investigation conforms with the Guide for the Care and Use of Laboratory Animals published by the US National Institutes of Health (NIH Publication No. 85-23, revised 1996). The Animal Ethic Review Committees of Zhejiang University approved all procedures.

Data analysis

In present study, the multiple biologic responses for ten combinations were used to fit an equation for simplex lattice model which then can predict the properties of all possible combinations. Data were shown as mean ± SD (n = 8). Statistical analysis were analyzed by anova at the level of p = 0.05. Graphs of these properties in the form of contour plots were developed with MiniTab Software (Minitab Inc., State College, PA). A multiple linear regression algorithm was used to generate model equation for representing relationship between the composition and the measured characteristics.

Results and Discussion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

QCAR models

Data in Table 1 were the training set used in present study. It is clear that the LAD ligation rats (Model group) have largest infarct rate of heart and MDA level, which indicated serious myocardial injury were induced by the ischemia. The treatment of three components PD, PT and OP individually cannot significantly decrease infarct size and MDA level. However, the combinations consisted of PT and OP (C8 and C9 group) can significantly attenuate the infarct size of heart. Moreover, the combinations of OP and PD (C5 and C6 group) shows better efficacy on serum MDA level than others.

Simplex lattice method was used to optimize the combination of three components from Shenmai formula. The proportion of PD (X1), PT(X2), and OP(X3) were chosen as the independent variables. Two different biologic activities, i.e. the infarction rate of heart and serum MDA level, were taken as responses (= [Y1, Y2]), respectively. The equation for simplex lattice model is described as follows:

  • image(4)

where Y is the dependent variable and bi is the estimated coefficient for the factor Xi. Based on the clinical dosage of Shenmai formula and results from our preliminary study, the daily administration of mixture was set at 150 mg/kg rat weight. Therefore, the dosage of three components PD, PT, and OP were calculated according to the proportion listed in Table 1. In order to simplify the computations, the actual concentrations of components were normalized based on the simplex lattice method so that the minimum concentration corresponds to zero and the maximum concentration corresponds to one.

Contour plots of independent variables were used to show how a response variable relates to three components based on a model equation. The multivariate regression function for infarct rate of heart was shown in eqn (5), and the function for MDA was shown in eqn (6).

  • image(5)
  • image(6)

The contour plot in Figure 3A also indicates that the combination of PT and OP lead to lower infarct rate, which suggests synergistic effect of OP and PT but not PD benefits to reduce the infarct area of heart during ischemia. This finding is in accordance with previous pharmacological report that PT leads to vascular angiogenesis which is of great benefit to ischemic heart, while PD has opposing active principles (40). Moreover, contour plot of MDA (Figure 3B) suggests that the combination of PD and OP with equal ratio may attenuate serum MDA level, a parameter showed the degree of lipid peroxidation injury of ischemic heart. From the contour figures, we can deduce that the combination of PD, PT, and OP in appropriate ratio will produce good efficacy in both infarct size and MDA level.

image

Figure 3.  Contour plots of cardioprotective effects for component combinations constructed by linear regression. (A) Infarct rate of Heart (%); (B) Serum MDA level.

Download figure to PowerPoint

Optimization of multicomponent combination from Shenmai formula

In Figure 3A,B, distinct relationships exist between the infarction area and MDA level with the restriction of constant dose. In clinical, decreased infarct size is more important than serum MDA level for patients with coronary heart disease. In order to attenuate infarct rate of heart and MDA level, the optimal composition of the multicomponent combination was selected by the linear model of the fused data.

Data fusion of two different biologic responses was performed by Derringer’s desirability function as stated in Methods Section. Using an extensive search algorithm with step size 0.001, the optimal solution for the integrated function for the proportion of PD, PT, and OP is 21.6%, 39.2%, and 39.2% respectively. Figure 4 shows the response surface of the fused dependent variable D. Thus, appropriate combination of three components from Shenmai formula was selected on the top-point of the response surface, which consists of PD (20%), PT (40%), and OP (40%) and was marked as V1.

image

Figure 4.  Response surface of the Derringer function. Spots V1, V2, and V3 represent the selected combinations for validation experiments.

Download figure to PowerPoint

The drug combination may produce a synergistic, an additive, or an antagonistic effect (41). In the present study, it is hard to determine interactions between three components from Shenmai formula belong to synergism or antagonism. Therefore, the assumption for the proposed QCAR-model is that their mixtures might cumulatively produce the overall additive effect. Peter Csermely et al. (42) suggested that multicomponent drug may produce higher effects or reduce range of side-effects in comparison with single-target drugs because of the interactions of multiple low-affinity bindings. System-oriented approach is regarded as ideal solution for the regulation of complex network systems, but it requires greater knowledge of the biologic features, and mechanisms of action for individual components and their combinations. Until recently, very few reports appeared in literature about optimization of combination aiming at several biologic properties or targets (43). When detailed principles of biologic network are still ambiguous, modeling QCAR from a controlled training set and searching optimal combination in a relative narrow composition-activity space is an alternative strategy for multicomponent drug design.

Validation of the computational results

To validate the computational results, additional experiment was performed using same animal model as mentioned in Data sets section. The decrease of infarct rate and MDA level induced by three combinations with distinct ratios of PD, PT and OP were assayed. In Figure 4, V1 represents the optimum combination calculated by proposed modeling and optimization approach. As shown in Figure 5, both combination V1 and V2 significantly attenuates the infarct rate of heart. But the efficacy of combination V2 (PD:OP = 1:1) and V3 (PD:PT = 1:1) are significantly lower than V1. Meanwhile, V1 also exhibits the best therapeutic effect on decreasing MDA level in three combinations. Therefore, the combination produced by the proposed QCAR-based method has potential to be applied to design multicomponent drug.

image

Figure 5.  Validation experiments for cardioprotective effects of three combinations on acute ischemic rats. Compared with myocardial infracted group (model), *p < 0.05; **, p < 0.01. Compared with V1 group, # represents p < 0.05; ## represents p < 0.01.

Download figure to PowerPoint

Conclusion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

In current study, we proposed a QCAR-based strategy for discovering optimal combinations of multiple components from TCM. The future success of this novel multicomponent drug-design paradigm will depend not only on a new generation of computer models to identify more correct multiple targets and their interactions but also on more reliable and high throughput biologic testing. In single-target compound design, the biologic data used for computation can be obtained by high-throughput assays on genes or proteins. But the major advantage of multicomponent drug lies on that its therapeutic effects through manipulating the cellular network and internal interaction, which suggests that multiple biologic responses should be observed in complex biologic system such as organ or animal models. The simplex lattice design reduces the requirement on the sample size for computational modeling without huge workload of biologic arrays. In present study, only ten groups of experiments were utilized to find optimal proportion of three components. It indicates that appropriate statistical experiment design and computational modeling could dramatically decrease the labor and cost in the multicomponent drug discovery.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Methods
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

This project was financially supported by the Chinese National Basic Research Priorities Program (No. 2005CB523402), the key grant from National Natural Science Foundation of China (No. 30830121), and China Postdoctoral Science Foundation (20081467).

References

  1. Top of page
  2. Abstract
  3. Methods
  4. Results and Discussion
  5. Conclusion
  6. Acknowledgments
  7. References