RESEARCH ARTICLE

Release modeling and comparison of nanoarchaeosomal,nanoliposomal and pegylated nanoliposomal carriersfor paclitaxel

Fatemeh Movahedi & Hasan Ebrahimi Shahmabadi &Seyed Ebrahim Alavi & Maedeh Koohi Moftakhari Esfahani

Received: 5 February 2014 /Accepted: 20 May 2014# International Society of Oncology and BioMarkers (ISOBM) 2014

Abstract Breast cancer is the most prevalent cancer amongwomen. Recently, delivering by nanocarriers has resulted in aremarkable evolution in treatment of numerous cancers. Lipidnanocarriers are important ones while liposomes andarchaeosomes are common lipid nanocarriers. In this work,paclitaxel was used and characterized in nanoliposomal andnanoarchaeosomal form to improve efficiency. To increasestability, efficiency and solubility, polyethylene glycol 2000(PEG 2000) was added to some samples. MTT assay con-firmed effectiveness of nanocarriers on MCF-7 cell line andsize measuring validated nano-scale of particles.Nanoarchaeosomal carriers demonstrated highest encapsula-tion efficiency and lowest release rate. On the other hand,pegylated nanoliposomal carrier showed higher loading effi-ciency and less release compared with nanoliposomal carrierwhich verifies effect of PEG on improvement of stability andefficiency. Additionally, release pattern was modeled usingartificial neural network (ANN) and genetic algorithm (GA).Using ANN modeling for release prediction, resulted in Rvalues of 0.976, 0.989 and 0.999 for nanoliposomal,pegylated nanoliposomal and nanoarchaeosomal paclitaxeland GA modeling led to values of 0.954, 0.951 and 0.976,

respectively. ANN modeling was more successful inpredicting release compared with the GA strategy.

Keywords Paclitaxel . Archaeosome . Liposome . PEG .

ANN . GN

Introduction

Recently cancer has become one of the major challenges. It isthe second leading cause of death in human societies [1]. Incase of cancer, cells are less active and unresponsive to apo-ptosis. In other words, they escape programmed cell death [2].Breast cancer is the most frequently diagnosed cancer inwomen and ranks second in cancer-related death. While theincidence has increased over the past decade, the mortalityrate of breast cancer has gradually declined. This improvedsurvival may stem from earlier detection as well as improvedtherapies [3]. Paclitaxel is a typical anticancer drug, with apotent antineoplastic activity against various types of solidtumors such as lung, ovarian and breast cancer [4, 5]; howev-er, its clinical application is limited yet due to lipophilicproperties and low water solubility and increased cytotoxicityand thus hypersensitivity reactions in case of using solventslike Cremophor EL. Additionally, non- targeted distribution ofdrug in living systems, results in reduction of efficiency andincreasing side effects and cytotoxicity [6]. An effective wayof overcoming the aforementioned challenges is encapsula-tion in nanocarriers such as common nanoliposome andnanoarchaeosome. Nanoarchaeosome is a novel and applica-ble liposomal carrier derived from Archaebacteria.

In this study, drug was entrapped in nanocarriers of lipo-some, archaeosome and pegylated liposome and differentcharacteristics of such nanocarriers like size, encapsulationefficiency, release pattern and cytotoxicity were evaluated.

F. MovahediSchool of Chemical Engineering, Iran University of Science andTechnology, Tehran, Iran

F. Movahedi :H. Ebrahimi Shahmabadi : S. E. Alavi (*) :M. Koohi Moftakhari Esfahani (*)Department of Pilot Nanobiotechnology, Pasteur Institute of Iran,Tehran, Irane-mail: s.ebrahimalavi@gmail.come-mail: maedehkoohi@gmail.com

S. E. Alavi :M. Koohi Moftakhari EsfahaniDepartment of Chemical Engineering, Science and Research Branch,Islamic Azad University, Tehran, Iran

Tumor Biol.DOI 10.1007/s13277-014-2125-4

In addition to the aforementioned issue, drug release wasevaluated using artificial neural networks (ANNs) and geneticalgorithms (GAs).

ANN is a mathematical model which tries to present arelationship between input and output recognizing substantialinterconnections using learning phase and neuron processors[7]. ANNs were inspired by human’s brain and neural systemincluding plenty of neurons. Such networks are able to learn,remember and link data [8].They are intelligent, dynamic andmodel-less networks based on experimental data and indepen-dent from any sort of postulation and rule [9]. Hidden layersprocess data received by input layer in order to represent themfor output layer. Each network is trained by some examplesand training results in learning. Learning process occurs incase of alteration in layer weights (coefficient) when thedifference between calculated and predicted value is tolerable.Such conditions make the learning process possible. Trainedneural networks are able to be applied for prediction of outputsof new set of data [8, 10]. Figure 1 demonstrates the schematicarchitecture of a network [11].

GA is a powerful optimizational tool inspired by naturalevolution. GAs are a good choice for predicting techniquesbased on regression. This algorithm uses biological tech-niques such as heredity and mutation in solving a problem[12]. GAs and ANNs are preferred to classical methods suchas response surface methodology for prediction and datafitting [13].

In this work, ANNs and GAs were used in order to predictrelease pattern of nanocarriers and thereupon, the mentionedmodels were compared.

Materials and methods

Materials

Phosphatidylcholine, cholesterol, paclitaxel, polyethyleneglycol 2000 (PEG 2000) and MTT (0.5 mg/ml) werepurchased from Sigma. Ethanol, Isopropanol, DMSO,pepton, yeast extract, KCl, HCl, Mg2Cl2⋅5H2O,MgSO4⋅7H2OandCaCl2⋅2H2O were obtained from Merck,RPMI-1640 from Invitrogen, and MCF-7 cells were sup-plied by Pasteur Institute of Iran.

Preparing nanoliposomes and loading the drug

Phosphatidylcholine and cholesterol (1:15) were dissolved in120 ml of 98 % ethanol (300 rpm, room temperature) andmixed with 13 mg of paclitaxel by magnetic stirrer (300 rpm,30 min, room temperature). Then solvent was evaporatedusing rotary evaporator (Heidolph, Germany). Resultant gelwas dissolved in 13 ml of physiological serum. For preparingpegylated nanoliposome, PEG 2000 was also added. All

compounds were sonicated (Bandelin Sonorex Digitec,60 Hz) for 5 min.

Preparing archaeosome and loading the drug

Archaeosomes were extracted according to the Bligh andDyer method [4]. Total polar lipids (TPL) were precipitatedusing cold acetone and resuspended and stored in chloroform/methanol [14].

Archaeosomes were prepared by hydrating extracted TPLin phosphate buffered saline (PBS) containing 10 mg ofpaclitaxel dissolved in 100 μl of DMSO. In order to uniformthe size of archaeosomes and increase drug loading efficiency,the solution was sonicated (Bandelin Sonorex Digitec, 60 Hz)for 10 min.

Determining size of nanocarriers

Average diameter of liposomal, pegylated liposomal andarchaeosomal paclitaxel was determined by Zeta sizer(Malvern Instruments Ltd).

Fig. 1 Schematic architecture of a network. Green boxes show inputlayer. There are three active layers (two hidden layers and output layerwhich are indicated by blue circles). Layers are fully connected andconnections to biases (red boxes) are showed by heavier lines (Terflothand Gasteiger 2012)

Tumor Biol.

Encapsulation efficiency

In order to determine amount of loaded drug, samples (nano-particles containing drug) were centrifuged at 4 °C and13,000 rpm for 30 min and supernatant was analyzed byspectrophotometric method (UV-1601PC Shimadzu) in227 nm to find amount of unloaded drug. Loading efficiencywas calculated using below formula:

Loading efficiency% ¼ Amount of encapsulated drug

Total amount of drug� 100

In vitro release study

Equal volumes of nanoarchaeosomal, nanoliposomal andpegylated nanoliposomal paclitaxel were poured in a dialysis

bag in 100 ml of saline buffer phosphate (pH 7.4) placed on amagnetic stirrer (25 h, 37 °C) to determine release pattern.Release in PBS buffer was measured at 227 nm by spectro-photometric method in certain time intervals within 24 h andthe percentage of released paclitaxel was obtained using drugstandard curve.

Cytotoxicity assay

Briefly, 100 μl of MCF-7 culture medium containing 10,000cells was poured in 96-well plate and incubated at 37 °C with5 % CO2. After 24 h, the supernatant was removed anddifferent concentrations of nanoliposomal, pegylatednanoliposomal and nanoarchaeosomal paclitaxel as well ascontrol samples were added to cells and incubated for another24 h. Afterwards, the supernatant was removed again, andMTT (0.5 mg/ml) solution was added and after 3 h, purplecolor of viable cells (due to formazan formation) was dis-solved in 100 μl isopropanol. Absorbance was measured in570 nm (Power Eave XS spectrophotometer) and IC50 wascalculated by Pharm program.

Modeling release of the drug

The first step in modeling by neural networks is nominatingsome effective factors as inputs as well as determining target.To specify the weight of layers, definite values are presentedto input layer and multiplied by the relevant layer. Aftershifting to next cell, all input amounts are summed and passedthrough activity function and as a result, output is defined.During training period, error is calculated using training algo-rithm and network weights are regulated according to magni-tude of error and a parameter called “rate of learning.” A final

Fig. 2 Schematic view of GA. First population is initialized and fitnessof each individual is evaluated and fittest ones are selected using geneticoperators such as crossover and mutation and next generation is formed.These steps are repeated iteratively (Terfloth and Gasteiger 2012)

Fig. 3 Percentage of releaseddrug for standard paclitaxelduring 42 h

Tumor Biol.

evaluation phase is essential to ensure accuracy and compre-hensiveness [15].

The way in which neural cells are connected in differentlayers, has resulted in various arrangements in ANNs. Multi-ple layer perceptron (MLP) is one of the most commonarrangements which is a standard combination of inputs,intermediates and outputs. The output of each unit of process-ing in each layer is shifted to next layer. The processor units offirst layer are linear but nonlinear neurons are used in hiddenlayers. Sigmoidal or hyperbolic tangential activity functionsare commonly used for such structures. In order to increase therate of learning, neurons of output layer are generally linear[16].

In this study a three-layer network with 20 neurons inhidden layers was used. Number of neurons in hidden layerwas designated considering mean square error (MSE) forpredicting released drug in comparison with experimentaldata. Feed-forward back propagation algorithm was used fornetwork training.

In a population of solutions, each individual is encoded bya chromosome. Each chromosome has a value of fitness whichcan determine its ability to survive. In this case, generation ofnew solutions is developed by genetic operators [11, 17]. Thismethod is based on “the survival of the fittest.” Generally,evolution starts from a population of randomly generatedindividuals. In each generation, the fitness of each individualis evaluated and the fittest candidates are selected to form nextgeneration (next iteration). The algorithm terminates by find-ing a solution meeting minimum criteria. Figure 2 illustratesthe schematic view of GA [11].

Modeling and statistical analysis

Results are presented as mean value±standard deviation (n=3).They were analyzed by one-way analysis of variance(ANOVA) using IBM Statistics SPSS software version 19. Topredict release pattern from nanocarriers, ANN and GA were

Fig. 4 Percentage of releaseddrug for nanoliposomal paclitaxelduring 42 h

Fig. 5 Percentage of releaseddrug for pegylated nanoliposomalpaclitaxel during 42 h

Tumor Biol.

applied. The ANN and GA programming were carried outusing MATLAB version 2009b (Mathworks, 2009).

Results and discussion

Determining the size of nanocarriers

The average diameter of nanoliposome, pegylatednanoliposome and archaeosome was 421.4 (polydispersityindex (PDI)=0.21) nm, 369.1 nm (PDI=0.13) and 521.4 nm(PDI=0.18), respectively. All sizes are validated in nano-scale. The size of nanoparticles is large compared with othersimilar works (about 200 nm) [18, 19], and probably such adifference in size has led to high level of encapsulation effi-ciency in this work. The smaller size in pegylated form is dueto interaction of different charges of liposome and PEG.

Encapsulation efficiency

Encapsulation efficiency—which was measured as describedabove— was determined to be 91.3±5.7 %, 95.2±6.3 % and99.1±3.7 % for nanoliposomal, pegylated nanoliposomal andnanoarchaeosomal paclitaxel in 26 h, respectively. A similarwork by Yang et al. [18] resulted in encapsulation efficiencyof at most 70 %. Thus, encapsulation efficiency is consider-ably high in all samples, and successful loading of drug ontonanoparticles is verified.

Nanoarchaeosomal carriers demonstrated highest encapsu-lation efficiency as well as lowest level of release. This proba-bly comes from the fact that archaebacterial lipids have a uniquestructure in comparison with other lipids [7]. They can beformed by the connection of glycerol and isoprenoid [14]. Sucha structure results in distinguished properties like stabilityagainst oxidation, pH and lipases [5]. Pegylated nanoliposomesshowed higher encapsulation efficiency and less release

compared with nanoliposomal carrier. This fact may verifyeffect of PEG on improvement of stability and efficiency.

In vitro release study

Release of paclitaxel in PBS buffer for standard drug andnano l i po soma l , p egy l a t ed nano l i po soma l andnanoarchaeosomal carriers during periods of 30 min and 1,1.5, 2, 3, 24, 25, 26, 27, 28, 29, 30, 40, 41, 42 h wasdetermined using standard curve (Figs. 3, 4, 5 and 6).

Released drug in PBS buffer after 42 h was 5.93 % fornanoliposomal, 5.33 % for pegylated nanoliposomal and0.161 % for nanoarchaeosomal carriers. Similar studies re-ported higher level of release [19].

Fig. 6 Percentage of releaseddrug for nanoarchaeosomalpaclitaxel during 42 h

Fig. 7 Predicted values for nanoliposomal paclitaxel using ANN incomparison with experimental data

Tumor Biol.

The low level of release in this work demonstrates the highlevel of stability in nanoparticle form and evidently presents thepossibility of sustained release using drug delivery systems.

Cytotoxicity assay

The IC50 values for nanoarchaeosomal, pegylatednanoliposomal, nanoliposomal and standard paclitaxel were

80.8±1.37, 79.8±2.9, 86.25±3.4 and 166.9±2.9 μg/ml, re-spectively. Apparently, there is a significant difference be-tween efficacy of Paclitaxel in nanoparticle form and standardform which confirms the potential of nanoparticles for Pacli-taxel delivery.

ANNs and genetic strategy-based model

A three-layer network with 20 neurons in hidden layer wasemployed to predict release pattern. Network training proce-dure was designed based on Levevberg–Marquardtbackpropagation algorithm. To predict released paclitaxel, V, tandCc were used for training, where V is the molecular volume,t is the releasing time, and Cc is the concentration of carrier.After the training phase, three other sets of t, V and Cc — notinvolved in the training phase — were used for network eval-uation. Purlin and Radbas transfer functions were used foroutput and hidden layers, respectively. Predicted and experi-mental release data are compared in Figs. 7, 8 and 9.

Fig. 9 Predicted values for nanoarchaeosomal paclitaxel using ANN incomparison with experimental data

Table 1 Statistical analysis of ANN modeling

Mean squared error(MSE)

R

Nanoliposomal paclitaxel 1.86×10−8 0.97604

Pegylated nanoliposomalpaclitaxel

1.53×10−23 0.98928

Nanoarchaeosomal paclitaxel 5.35×10−10 0.99974

Fig. 10 Predicted values for nanoliposomal paclitaxel using GA incomparison with experimental data

Fig. 8 Predicted values for pegylated nanoliposomal paclitaxel usingANN in comparison with experimental data

Tumor Biol.

The R values for predicting release of nanoliposomal,pegylated nanoliposomal and nanoarchaeosomal paclitaxelare 0.97604, 0.98928 and 0.99972, respectively. Table 1shows the statistical analysis of the model.

In another phase of this work, genetic strategy was usedto train the network. This strategy uses GA to specify a planfor weight of inputs. In other word, this strategy evaluates

plans to find the best one for training. Figures 10, 11 and 12represent a comparison between experimental and predictedrelease data.

Table 2 illustrates the statistical analysis for modeling byGA strategy.

Conclusion

In this work, we synthesized nanoliposomal, pegylatednanoliposomal and nanoarchaeosomal paclitaxel successfully.The cytotoxicity effect of all formulations confirmed effec-tiveness of nanocarriers. Such an improvement can come fromthe amphiphatic nature of liposomes which make them able toencapsulate a lipophilic drug like paclitaxel as well as carryingit in aqueous media. Results showed that most of the drug wasencapsulated into the carriers. Measuring the mean diameterof synthesized carriers validated the size in nano-scale, andpegylated nanoliposomal carrier was the smallest one. Thenanoarchaeosomal carrier demonstrated the highest encapsu-lation efficiency as well as the lowest level of release. Inaddition, release pattern was modeled using ANN and GAstrategy.

MSE was negligible for all formulations in both models;however, the highest values of R were obtained fornanoarchaeosomal carriers. It should be added that ANNmodeling demonstrated higher values of R. Consequently,applying the GA strategy cannot be a perfect choice.

All in all, applying both strategies can result in acceptableprediction of drug release but ANN can be the preferred one.Prediction can be more efficient using re-training by extensivesets of data. Additionally, prediction of release for othernanomedicines is possible using this method which can pre-dict different factors affecting efficiency of nanomedicines ina fast and precise manner. Such modeling strategies can makeopportunity of innovative characterization and optimizationfor design and development of nanomedicines parallel withexperimental trials to find further clinical potentials for cancertreatment.

Conflicts of interest NoneFig. 12 Predicted values for nanoarchaeosomal paclitaxel using GA incomparison with experimental data

Table 2 Statistical analysis of GA modeling

Mean squared error(MSE)

R

Nanoliposomal paclitaxel 2.18×10−10 0.95463

Pegylated nanoliposomalpaclitaxel

1.01×10−12 0.95129

Nanoarchaeosomal paclitaxel 2.11×10−10 0.97582

Fig. 11 Predicted values for pegylated nanoliposomal paclitaxel usingGA in comparison with experimental data

Tumor Biol.

References

1. Guo J, Bourre L, Soden DM, O'Sullivan GC, O'Driscoll C. Can non-viral technologies knockdown the barriers to siRNA delivery andachieve the next generation of cancer therapeutics? Biotechnol Adv.2011;29:402–17.

2. Blagosklonny MV. A node between proliferation, apoptosis, andgrowth arrest. Bioessays. 1999;21:704–9.

3. Rostas JW, Dyess DL. Current operative management of breastcancer: an age of smaller resections and bigger cures. Int J BreastCancer. 2012;2012:516417.

4. Bligh EG, Dyer WJ. A rapid method of total lipid extraction andpurification. Can J Biochem Physiol. 1959;37:911–7.

5. Brasnjevic I, Steinbusch HW, Schmitz C, Martinez-Martinez P.European NanoBioPharmaceutics Research Initiative. Delivery ofpeptide and protein drugs over the blood–brain barrier. ProgNeurobiol. 2009;87:212–51.

6. Dorr RT. Pharmacology and toxicology of Cremophor EL diluent.Ann Pharmacother. 1994;28:11–4.

7. Sprott G. Structures of archaebacterial membrane lipids. J BioenergBiomembr. 1992;24:555–66.

8. Dayhoff JE. Neural network principles. 1st ed. Prentice Hall: USA;1990.

9. Neocleous C, Schizas C. Artificial neural network learning: a com-parative review. 1st ed. Berlin: Springer; 2002.

10. Richard BD, Liu YA. Neural network in bioprocessing and chemicalengineering. 1st ed. USA: Academic Press; 1996.

11. Terfloth L, Gasteiger J. Neural networks and genetic algorithms indrug design. Genomics Suppl. 2012;6:102–8.

12. Mukhopadhyay DM, Balitanas MO, Farkhod AA, Jeon SH,Bhattacharyya D. Genetic algorithm: a tutorial review. Int J GridDistrib Comput. 2009;2:25–32.

13. Graupe D. Principles of ANN. 2nd ed. Singapore: World ScientificPublishing; 2007.

14. Krishnamachari Y, Geary SM, Lemke CD, Salem AK. Nanoparticledelivery systems in cancer vaccines. Pharm Res. 2011;28:215–36.

15. Haykin S. Neural networks: a comprehensive foundation. 2nd ed.Prentice Hall: USA; 1998.

16. Laugier S, Richon D. Use of artificial neural networks for calculatingderived thermodynamic quantities from volumetric property data.Fluid Phase Equilib. 2003;2:247–55.

17. Rizkalla N, Hildgen P. Artificial neural networks: Comparison of twoprograms for modeling a process of nanoparticle preparation. DrugDev Ind Pharm. 2005;31:1019–33.

18. Yang T, Cui FD, Choi MK, Cho JW, Chung SJ, Shim CK, et al.Enhanced solubility and stability of PEGylated liposomal paclitaxel:in vitro and in vivo evaluation. Int J Pharm. 2007;338:317–26.

19. Zhang JA, Anyarambhatla G, Ma L, Ugwu S, Xuan T, Sardone T,et al. Development and characterization of a novel Cremophor ELfree liposome-based paclitaxel (LEP-ETU) formulation. Eur J PharmBiopharm. 2005;59:177–87.

Tumor Biol.