February 2013 125Chem. Pharm. Bull. 61(2) 125–133 (2013)

© 2013 The Pharmaceutical Society of Japan

Poly(dl-lactide-co-glycolic acid) Nanoparticle Design and Payload Prediction: A Molecular Descriptor Based StudySuvadra Das, Partha Roy, Md Ataul Islam, Achintya Saha, and Arup Mukherjee*Department of Chemical Technology, University of Calcutta; 92 A.P.C. Road, Kolkata-700009, West Bengal, India.Received May 23, 2012; accepted October 29, 2012; advance publication released online November 29, 2012

Polymer nanoparticles are veritable tools for pharmacokinetic and therapeutic modifications of bioactive compounds. Nanoparticle technology development and scaling up are however often constrained due to poor payload and improper particle dissolution. This work was aimed to develop descriptor based computational models as prior art tools for optimal payload in polymeric nanoparticles. Loading optimization experiments were carried out both in vitro and in-silico. Molecular descriptors generated in three different platforms DRAGON, molecular operating environment (MOE) and VolSurf+ were used. Multiple linear regression analysis (MLR) provided computation models which were further validated based on goodness of fit statis-tics and correlation coefficients (DRAGON, R2=0.889, Q2=0.657, R2

pred=0.616; MOE, R2=0.826, Q2=0.572, R2

pred=0.601; and VolSurf+, R2=0.818, Q2=0.573, R2pred=0.653). Pharmacophore space modeling studies were

carried out in order to understand the fundamental molecular interactions necessary for drug loading in poly(dl-lactide-co-glycolic acid). The space modeling study (R2=0.882, Q2=0.662, R2

pred=0.725, Δcost=108.931) indicated that hydrogen bond acceptors and ring aromatic features are of primary significance for nanopar-ticle drug loading. Results of in vitro experiments have also confirmed the fact as a viable prognosis in case of nanoparticle payload. Polymeric nanoparticles payload prediction can therefore be a useful tool for wider benefits at the preformulation stages itself.

Key words poly(dl-lactide-co-glycolic acid); molecular descriptor; nanoparticle

The drug discovery process is severely constrained in the current era because of steep rise in R&D expenditure and at the same time only a few practicable chemical entities are being reported.1) Complexity arising out of drug toxicity and disease management has however remained unabated. Dis-covery process therefore has seen a paradigm shift in favor of pharmacokinetic modulation of known chemical entities rather than pharmacodynamic developments.2) A search in favor of purified plant products has also been renewed be-cause of inherent low toxicity and easier availability.3) Of late, rapid developments in nanomedicines as platform technology have provided further scope in pharmacokinetic modulations of known chemical entities. Polymer associated nanoparticle engineering is known to enable drug molecules remain longer in systemic circulation or help target them to cellular sites. Modulations of size, charge and surface chemistry are known to assist in such effects.4) Enhanced kinetic targeting is pass-able for nanomedicines particularly in tumors or in sites of inflammation.5,6) Remarkable success in such therapy have already been achieved with polymeric nanoparticles.7)

Despite many advantages polymeric nanoparticles are faced with overwhelming challenges in preparative stages.8) Preparation of polymeric nanomedicines requires multiple trial and error runs leading to high cost and loss of precious drug molecules. Specific successes are often unpredictable and up-scaling practices difficult. Drug loading is generally unpredictable in all cases including known chemical entities and plant drug molecules. Diverse physicochemical interac-tions are considered as determinants in successful polymeric nanoparticle drug loading.9)

Quantitative structure performance relationship (QSPR) tools have earlier been used in drug discovery and chemi-cal toxicology.10,11) Modeling tools for accurate assessment of

physicochemical properties and molecular interactions were known in pharmaceutics.12,13) Application of in-silico tools in nanomedicine developments is relatively recent. Costache et al. have used modeling tools to understand drug polymer interactions in polyethylene glycol (PEG)-tyrosine copolymer nanospheres.14) One interesting recent study has used descrip-tor based modeling tools for nanostructure biological property relationships studies.15) Simulation tools were very recently experimented in design of targetable nanoparticles16) and an excellent review to extrapolate QSPR techniques as a predic-tive tool for nanomaterial properties has just been published.17)

Herein, we intend to develop and evaluate in-silico models as prior art tools for nanoparticle payload predictions. Poly(dl-lactide-co-glycolic acid) (PLGA) was taken as the polymer material. PLGA is the most widely used polymer for nanopar-ticle drug delivery and is also approved by U.S. Food and Drug Administration (FDA).18) Descriptor based tools were used to work out the fundamental relations responsible for successful molecular payload. Physicochemical interactions that are important in drug loading were further studied fol-lowing probe mapping techniques. Parallel in vitro experiment was followed with known chemical entities that associate bio-pharmaceutical limitations. Computational design of nanopar-ticle drug delivery devices is seen as a substantial solution and prior-hand tool particularly for selection of loadable molecules and polymeric materials.

ExperimentalMaterials Silybin (Sb), Polyvinyl alcohol (PVA), PLGA

(50 : 50, molecular weight (MW) 40000–75000), and dialy-sis tubing D9652 (MW cut off 12400) were purchased from Sigma Aldrich (St. Louis, MO, U.S.A.) and Andrographolide (AG) was from Natural Remedies (Bangalore, India). All solvents and water used in in vitro experiments were HPLC grade and were purchased from E. Merck or Spectrochem

Regular Article

* To whom correspondence should be addressed. e-mail: arupm1234@gmail.com

The authors declare no conflict of interest.

126 Vol. 61, No. 2

(Mumbai, India).Apparatus Standard Borosil® glassware (Mumbai, India)

was used for preparation and analysis experiments. A Vibra cell VCX 750 W sonicator (Sonics, Newtown, U.S.A.), a ho-mogenizer model TH 02 (Omni International, Kennesaw, U.S.A.), a precision balance Mettler-Toledo AL54 0.00001-g (Mettler, Columbus, OH, U.S.A.), Tarsons Accupipets (Tar-sons, Calcutta, India), and Himac CS120GHXL ultracentri-fuge (Hitachi Koki, Tokyo, Japan) were used in preparative processes. Chemical analyses were carried out in a Waters dual pump HPLC model 515 (Waters, New Castle, DE, U.S.A.). A zetasizer Nano ZS (Malvern Instruments, Mal-vern, U.K.), a Nanoscope 3A atomic force microscope (Veeco, Plainview, NY, U.S.A.) were used for most particle character-ization and imaging experiments.

Nanoparticle Preparation Emulsion and solvent evapo-ration technique was followed throughout keeping all pa-rameters constant except the loadable compound Sb or AG. Nanoparticle preparation process was as detailed by us else-where.19) Initial mass ratio for PLGA polymer and the loadable drug was also kept constant. Briefly, PLGA was dissolved in organic phase alongside AG or Sb and the organic phase was emulsified under sonication (20 kHz, 750 w) for one minute in an aqueous solution containing 4% w/v PVA. The resultant oil/water emulsion was further homogenized at 30000 rpm for 1 h under external cooling in ice water. The solvent evapora-tion was continued for 12 h over a magnetic stirrer at room temperature and nanoparticles (AGnp or Sbnp) were harvested by ultracentrifugation at 30000 rpm for 30 min at 4°C. These were washed in water, re-centrifuged and preserved in vacu-um desiccators at 4°C until further evaluations.

Nanoparticle Characterization Nanoparticles (AGnp, Sbnp) were characterized for their particle size, poly dispersity index (PDI) and zeta potential in Zetasizer nano ZS. Particle size was measured against a 4mw He–Ne laser beam 633 nm, and back scattering angle of 173° and the PDI was also re-corded. Zeta potential was determined on the basis of electro-phoretic mobility under an electrical field using Smoluchosky equation. The morphology examination of the nanoparticles of both AGnp and Sbnp types were carried out under an atomic force microscope in tapping mode using RTESP tip at 267–328 kHz resonance frequency and a scan speed of 1.2 Hz.

Nanoparticle Loading AG was analysed at 228 nm in a HPLC set up. The mobile phase for analysis was aceto-nitrile–0.1% (v/v) phosphoric acid in water (40 : 60, v/v), flow rate 1 mL/min.19) A peak area (y) vs. concentration (x) graph was first drawn from different known concentrations of AG. Analyte mass originally taken and that in supernatant after nanoparticles preparation were determined separately from the HPLC peak area observations.

In case of Sb the HPLC PDA detector (2996, Waters, U.S.A.) was set at 288 nm and a mixture of 85% phosphoric acid–methanol–water (0.5 : 46 : 64, v/v) was used as the mobile phase.20) The solvent flow rate was 1 mL/min. Both analysis were carried out using C18 column (Supelco, Bellefonte, PA, U.S.A.), 25×4.6 mm.

Percentage drug loading was determined from triplicate ex-periments and were represented as

molecular loading (%)

mass of drug originally taken mass of drug in supernatantmass of drug originally taken

100

=

×

Computational Study Compiling a diverse molecular datasets with known mass loading was complex due to varied polymer choices, different experimental conditions used and also the preferences in reporting mass loading. Manual, text-based searches were undertaken from a range of data source including PubMed, Web of Sciences, ACS, Scopus, Science Direct and Wiley Publication Resource Centers under insti-tutional subscriptions. Logarithmic form (log10) of percentage mass loading was used as the response parameter. A dataset of 22 compounds was build based on literature survey and our laboratory experiments. The assimilated data set was random-ly divided into training set (Tr) and test set (Ts) for modeling and validation studies. ChemDraw, was initially used to gen-erate the 2-dimensional (2-D) structures of the molecules and further 3-D structures were generated for inspections using Chem3DUltra (Chem Office 2002 version 7.0, U.S.A.). Mo-lecular geometries were optimized in MOPAC (MOPAC 2000 version 7.0, U.S.A.) using AM1 semi empirical method and used further for computation of different model descriptors.

Descriptors One objective of this study was to capture the most informative descriptors set and finally develop a hy-pothesis for high drug mass loading in PLGA nanoparticles. Three independent microscopic descriptor machines namely DRAGON, molecular operating environment (MOE) and VolSurf+ were used throughout.

DRAGON Descriptors: All 0D, 1D, 2D and 3D descrip-tors generated by DRAGON software (DRAGON version 5.5, U.S.A.) were explored to accommodate wider variety of types. 0D descriptors provided the molecular constitutional propertis,21) 1D descriptors provided functional group counts and atom-centered fragments,22) 2D descriptors extended topo-logical descriptors, connectivity indices and eigen value based indices23) and 3D descriptors provided different molecular geometry functionalities.24–27) A total of 1504 number of de-scriptors were generated through DRAGON. Descriptors with invalid values, with too many zeros and with small standard deviation (less than 0.5) were rejected in further applications.

MOE Descriptors: Two classes of molecular descriptors, namely 2D and 3D molecular descriptors (“i3D” internal co-ordinate dependent and “x3D” external coordinate dependent) were calculated using the molecular operating environment modeling program (MOE version 2007.09, Canada). The 2D molecular descriptors are the numerical properties evalu-ated from the connection tables representing a molecule and include physical properties, subdivided surface areas, atom counts, bond counts, Kier and Hall connectivity28) and κ shape indices,29,30) adjacency and distance matrix descriptors containing BCUT and GCUT descriptors,31–35) pharmacophore feature descriptors, and partial charge descriptors (PEOE de-scriptors).36) Different 3D descriptors37) include a number of quantum chemical descriptors, such as the molecular dipole moments, heat of formation, ionization potentials, highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energies calculated using the AM1 semi-empirical method. The total number of MOE descriptors finally considered was 201.

February 2013 127

VolSurf+ Descriptors: VolSurf+ descriptors were calcu-lated using VolSurf+ (VolSurf+ version 1.0.4, U.K.) in win-dows platform. VolSurf+ was used to generate 2D molecular descriptors from 3D molecular interaction energy grid maps.38) The basic idea in VolSurf+ is to compress the information present in 3D maps into well defined 2D numerals. These descriptors quantitatively characterize the size, shape, polarity and the hydrophobicity of molecules. GRID probes used in the calculation of the VolSurf+ descriptors were, the water probe (OH2), the hydrophobic probe (DRY), the carbonyl oxygen atom probe (O) and amide probe (N1). Grid spacing was set to 0.5 Å and a total 128 descriptors were derived for further applications.

Multiple Linear Regression (MLR) Analysis Linear correlation models were derived through standard and forward stepwise regression techniques in STATISTICA (Statistica version 7.1, U.S.A.). Descriptor inter-correlations were further cross checked and the final MLR equation was developed ac-commodating only the principle non-correlated descriptors. All data were statistically analyzed through computation of R2 (correlation coefficient), F (variance ratio) with df (degrees of freedom), s (standard error of estimate). The final selected model was cross-validated through leave-one-out (LOO) technique39) applying Pressav (average of absolute value of predicted residuals), SDEP (standard deviation of error of pre-dictions) and Q2 (cross-validated variance) as model validation tools. The analysis was performed in HP xw6600 work station with 4GB RAM and 1TB hard disk.

Space Modeling CATALYST40) (CATALYST version 4.11, U.S.A.) was used to understand the fundamental molecular forces important for molecular payload in polymeric particles. A pharmacophore model consists of a 3D arrangement of a collection of chemical features necessary for the activity of the ligands. The input chemical features considered for this study were hydrogen bond (HB) acceptor (a) and donor (d), hydrophobic (p), and ring aromatic (r). Different control pa-rameters used for hypothesis generation were spacing, uncer-tainty and weight variation. Spacing represented the minimum inter-features distance that may be allowed in the resulting hy-pothesis. Each parameter in the generated hypothesis implies an order of magnitude of the compound’s activity. The level to which it is explored by the hypothesis generator is controlled by the weight variation parameter. The uncertainty parameter reflects the degree of confidence and tolerance factor of the biological data. Compounds were subjected to internal strain energy minimization and conformational analysis. The costs of null and fixed hypotheses were calculated. A greater differ-ence between the fixed and the null costs is an indicator for lower chance correlations. The minimum difference between the total and null costs was taken as 60 bits for hypothesis optimization. The quality of hypothesis was further adjudged through a cross-validation technique using CatScramble based on Fischer’s randomization test, where the activity data were randomized within a fixed chemical data set.

In-Vitro Release and Modeling Payload release and

release kinetic modeling were studied for prepared nanoparti-cles in order to relocate the loaded molecules. AGnp and Sbnp equivalent to 2 mg of AG and Sb payload were suspended in 1 mL of phosphate buffer (100 mm, pH 7.4) and transferred separately in dialysis bags. The dialysis bags were then placed in glass vials containing 10 mL phosphate buffer. Both were kept in an incubator shaker at 37°C and 50 rpm. At prede-termined time intervals, 5 mL of release medium were with-drawn for analysis. Withdrawn samples were compensated by fresh buffer to maintain the sink condition. The amount of AG and Sb released per timed sample were estimated in HPLC method with necessary corrections for the dilution factors. The in-vitro release studies for all preparations were com-pleted in triplicate and the cumulative percentage release over time was plotted.

Results and DiscussionPLGA Nanoparticles A generalised emulsion and sol-

vent evaporation technique was followed throughout for nanoparticulation and the final particle characteristics were listed in Table 1. Nanoparticle loading percentage and the zeta potentials are influenced by the polymer characteristics and the loadable molecule used.41) Drug loading and the particle surface charge were different in case of AGnp and Sbnp due to varied polymer payload and physicochemical interactions. Both Sb and AG are hydrophobic compounds and have a close proximity in solubility characteristics. Large differences in nanoparticles loading percentage indicated that solubility fac-tors are not the only parameters responsible for higher mass loading in PLGA nanoparticles. Uniform surface morphology is a characteristic for PLGA nanoparticles. Both AGnp and Sbnp appeared spherical and of smooth surface when studied under atomic force microscope (Fig. 1).

Effectiveness of a drug delivery device is directly depen-dent on its drug payload. A superior molecular mass payload at the preparative stages not only reduces loss but can also extend a more effective delivery device. The mass payload in nanoparticles was determined in HPLC by measuring the amount of drug originally taken and the mass of non-entrapped drug present in the supernatant accumulated after centrifugation and washings. The standard graph equation for

Fig. 1. AFM Study of (A) AGnp and (B) Sbnp in Tapping Mode

Table 1. Nanoparticle Characteristics: PCS Studies and Molecular Loading Percentage

Nanoparticle types Particle size (nm) Zeta potential (mV) PDI Molecular loading (%)

AGnp 173±12.60 −34.8±1.40 0.229±0.01 80.0±3.50Sbnp 187±14.32 −28.4±1.30 0.234±0.01 25.0±1.20

128 Vol. 61, No. 2

AG analysis was y=28051x+ 51747, R2=0.9979 and that for Sb was y=25750x+30567, R2=0.9992. The HPLC peak area was denoted as y and x represented the analyte concentration. Nanoparticle loading observed for AG was 80% while that in case of Sb was very less (Table 1). This was considered pos-sible due to multiple molecular interactions during formation of loaded nanoparticles. Zeta potential of nanoparticles was negative due to biopolymer PLGA but was widely different in both cases.

In-Silico Studies One objective of this work was to gen-erate a relationship between the structural parameters and their corresponding mass payload in designated biopolymer PLGA. In-silico model was developed to understand, correlate and foretell payload in nanoparticles. Descriptor based MLR approach perfected in molecular modeling and drug design was used as a platform technology.42)

A survey through different literature sources revealed a range of drug loaded polymeric nanoparticles varying in pro-cess technology and selection of polymer types. PLGA 50 : 50 is however the only polymer approved by U.S. FDA for nano-medicine applications so far. Working data set was selected based on the uniformity of nanoparticle process technology and the polymer choice PLGA.

Altogether 22 compounds reported in literature43–59) in-cluding those reported from our laboratory were used for generation of analysis models (Table 2). Percentage mass loading was considered the property and used in logarithmic form. Enlisted compounds (Table 2) provided a wide range of chemical structures and a range of drug mass load in PLGA. These were used to develop minimum energy charge neutral-ized conformers for further studies. Molecular and atomic descriptors were generated separately on DRAGON, MOE and VolSurf+ platform.

The DRAGON platform developed a 22×1504 matrix while MOE platform provided a 22×201 matrix and VolSurf+ pro-vided a 22×128 matrix. Each matrix was fitted for a MLR equation in STATISTICA. The final reduced model was cross validated in each case and similar findings were reported else-where.42,60,61) Descriptor inter-correlations were cross checked and the final MLR equations were developed accommodating only the non-correlated descriptors. The best linear QSPR models for drug loading in PLGA in three different descrip-tor platform were depicted in Table 3. Log of percentage mass loading represented as log10A. Statistical significance of each regression equation was evaluated by the statistical parameters like R2 and F. The predictive power of the model was judged from Q2 values, SDEP and PRESSavg through LOO method. Each model was again validated through a test set estimation of R2

pred and the statistical qualities of models were presented in Table 4.

Among 1504 DRAGON descriptors after several permu-tations the optimum model generated with five descriptors (Table 3) where Mor29u is derived from infrared spectra simulation and represented the 3D structure of a molecule by a fixed number of variables irrespective of number of atoms in the molecule. GATS5m is a distance type function. C-019 is among the atom-centered fragments where atom types were classified in relation to their structural environment. T(N…O) depicted the topological features of the atom which described the compound similarities in a quantitative manner. MATS2m is the spatial autocorrelation molecular descriptor, related to

the atomic property like atomic mass.62)

The selected descriptors in the final model generated with MOE descriptor were E_strain, Reactive, SMR_VSA4, SMR_VSA7 and MNDO_HF. E_strain is among the potential energy descriptors which signifies the non-covalent interac-tion potential energy accountable for the loading affinity of the drug to the considered polymer. Reactive is related to the molecular property, SMR_VSA4, SMR_VSA7 are the descrip-tors based on approximate accessible van der Waals surface area calculation for each atom along with molar refractivity. MNDO_HF is related to heat of formation.62)

In VolSurf+ LgS10 indicates the importance of solubility parameter of drug molecule at pH10, WO6 is the hydrogen bond donor descriptor computed at 6 kcal/mol energy level. DD4 denotes the difference between the maximum hydropho-bic volume and the hydrophobic volume of the 3D structure of the drug molecule. DRACAC is the 3D pharmacophoric descriptor based on the triplets of pharmacophoric point de-scriptor.

The independent variables used in these 2D QSPR models were not inter-correlated (R<0.5). The observed vs. predicted mass loading (logarithmic form) as per DRAGON, MOE and VolSurf+ model were presented in Fig. 2.

Pharmacophore Space Modeling Study In order to further understand the specific payload polymer interactions in PLGA nanoparticles, a receptor independent space mod-eling study was performed with the dataset using hypogen algorithm of CATALYST. The best hypothesis was derived on the basis of highest cost difference (Δcost=108.931), low root mean square division (rmsd=1.203) and the best correlation (R2=0.882). The mapped pharmacophoric features adjudicated were hydrogen bond acceptor in three different zones and one hydrophobic feature in 3D space. The quality of the hypothe-sis generated was further decided through a cross-validation technique using Fischer’s randomization test at 95% confi-dence, where the activity (Y-randomization) data were ran-domized within the chemical data set and the hypogen process was initiated to explore possibilities of other hypothesis of predictive values. Further the value of total cost of the model 96.625 was closer to the fixed cost of 80.363, which further suggested closeness to the ideal hypothesis. The critical inter feature distance imparting drug loading in PLGA in 3D space were depicted in Fig. 3. The most active compound of the test set mapped all the features in the model hypothesis.

In-silico predictive mass loading for AG in PLGA was higher among the test compounds while that in case of Sb was much lower. These observations correlated well with the in vitro experiments for Sb and AG mass loading in PLGA nanoparticles. In-silico modeling therefore can extend a reasonable tool for selection of drug and polymer in design of nanoparticle drug delivery devices in the preformulation stages itself.

In-Vitro Drug Release and Modeling Studies Release studies of hydrophobic bioactives from polymeric nanodevices often encounter a number of challenges.63) Solubility of AG in water at 37°C is 3 µg/mL and that for Sb was 0.4 mg/mL. However solubility alone may not be the determinants for success in nanoparticle drug loading and release kinetics. Dis-solution and in vitro release was studied in dialysis bags and molecular leaching was retraced in HPLC experiments. Initial drug release pattern from both AGnp and Sbnp displayed

February 2013 129

Table 2. List of the Compounds Selected for Generation of QSPR and Pharmacophore Models for Prediction of Mass Loading in PLGA (Cont’d)

Compound name Structure Max attainable mass loading (%) Log (%mass loading) (experimental property)

Haloperidol 81.00 1.908

Triclosan 85.12 1.930

Clonazepam 15.70 1.196

Ellagic acid 61.56 1.789

Estradiol 62.24 1.794

Ropivacaine 3.80 0.580

Paclitaxel 43.10 1.634

Cisplatin precursor 18.40 1.265

130 Vol. 61, No. 2

Table 2. List of the Compounds Selected for Generation of QSPR and Pharmacophore Models for Prediction of Mass Loading in PLGA (Cont’d)

Compound name Structure Max attainable mass loading (%) Log (%mass loading) (experimental property)

Camptothecin 100.00 2.000

Risperidone 90.00 1.954

5-Flurouracil 15.00 1.176

Quercetin 1.36 0.134

Xanthone 33.00 1.519

RU58668 97.60 1.989

Mitoxantrone 90.95 1.959

Procaine dihydride 44.10 1.644

February 2013 131

Table 3. Collection of Empirical Descriptor Set and Validated Models in QSPR Study

Modelling source

Initial number of descriptors generated

Descriptors selected in MLR analysis Best fit linear QSPR models

DRAGON 1504 Mor29u,GATS5m,C-019,T(N…O),MATS2m

log10A= −0.680(±0.269) Mor29u −1.044(±0.196) GATS5m −0.293(±0.107) C-019 +0.003(±0.001) T(N…O) +2.034(±0.900) MATS2m +2.614(±0.217)

MOE 201 E_Strain,Reactive,SMR_VSA4,MNDO_HF,SMR_VSA7

log10A= 0.044(±0.009) E_Strain −0.609(±0.234) Reactive −0.044(±0.011) SMR_VSA4 +0.004(±0.001) MNDO_HF −0.007(±0.002) SMR_VSA7 +1.735(±0.214)

VolSufr+ 128 LgS10,WO6,DD4,DRACAC

log10A= −0.117(±0.022) LgS10 −1.171(±0.287) WO6 −0.214(±0.081) DD4 −0.013(±0.006) DRACAC +1.623(±0.131)

Table 2. List of the Compounds Selected for Generation of QSPR and Pharmacophore Models for Prediction of Mass Loading in PLGA (Cont’d)

Compound name Structure Max attainable mass loading (%) Log (%mass loading) (experimental property)

Letrozole 43.00 1.633

Rivastigmine 74.50 1.872

Sperfloxacin 60.60 1.782

Andrographolide 80.00 1.903

Silymarin 25.00 1.398

Curcumin 87.00 1.940

132 Vol. 61, No. 2

significant burst effect due to surface adsorbed molecules in PLGA nanoparticles. The later part of the release was attrib-uted to chain scission and hydrolytic erosion of the PLGA ma-trix.64) Amount of drug entrapped in nanoparticles often affect the release characteristics.65) AGnp released 37% of payload in 24 h as compared to 24% release in case of Sbnp over the same time period. AGnp displayed a sustained release pattern up to 720 h and 80% of the payload could be accounted in the study period (Fig. 4).

Korsemeyer–Peppas model (Eq. 1) was used to describe

the drug release mechanism from nanoparticles over a time range.66)

/ = nt αM M Kt (1)

where Mt, mass released at time point ‘t,’ Mα, total mass load, n is the release exponent and K is the constant accommodat-ing the structural and geometrical features. The n and k value recorded for both AGnp and Sbnp were represented in Table 5. The n values near 0.3 indicating that the release mecha-nism was not a fickian diffusion.67) Similar observations were reported earlier as a result of simultaneous PLGA polymer disintegration.19)

ConclusionDescriptor based QSPR models were developed for pre-

diction of drug payload in PLGA nanoparticles. Molecules expressing strong hydrogen bond acceptors, ring aromatic features and a distance geometry are predictable for higher percentage nanoparticle payload in PLGA biopolymer. The success of models generated was marked by oneness of the results. Parallel in vitro synthesis for AG and Sb in PLGA corroborated well with the in-silico results. New descriptor based QSPR model will be a beneficial tool in polymeric nanoparticle synthesis.

Acknowledgements Financial assistance to Mrs. Suvadra Das and Md Ataul Islam from the Council of Scientific and

Table 5. Korsemeyer–Peppas Release Kinetics for Nanoparticles

Nanoparticle typeKorsemeyer–Peppas model parameters

n Value K Value

AGnp 0.33 0.09Sbnp 0.30 0.09

Table 4. Statistical Parameters for QSPR Models

ModelsCorrelation statistics (training set) Prediction statistics (training set) Prediction statistics (test set)

1ntr2R2 3F 4df 5s 6PRESSavg

7SDEP 8Q2 9nts10R2

pred11sp

DRAGON 16 0.889 16.066 5,10 0.221 0.270 0.310 0.657 6 0.616 0.200MOE 16 0.826 9.525 5,10 0.302 0.301 0.378 0.572 6 0.601 0.198

VolSurf+ 16 0.818 13.465 4,12 0.278 0.268 0.361 0.573 6 0.653 0.2041 No. of compounds of training set; 2 Correlation coefficient; 3 Variance ratio; 4 Degree of freedom; 5 Standard error of estimate; 6 Average of absolute value of predicted re-

siduals; 7 Standard deviation of error of predictions; 8 Cross-validated variance; 9 No. of compounds in test test; 10 Predictive R2; 11 Standard error of prediction.

Fig. 2. Observed vs. Predicted % Mass Loading (Logarithmic Form) in 2D QSAR Models

Fig. 3. Mapped Pharmacophore Features (Hydrogen Bond Acceptor (a), Hydrophobic (p)) and Critical Inter Feature Distance Important for Molecular Loading in PLGA

Fig. 4. In-Vitro Release Studies of AGnp and Sbnp

February 2013 133

Industrial Research (CSIR) and Mr. Partha Roy from Indian Council of Medical Research (ICMR) are gratefully acknowl-edged. Partial support from the Department of Biotechnology, Govt. of India is also acknowledged. The authors are thank-ful to Mr. Anirban Chakraborty for his assistance in Atomic Force Microscopy studies and to Dr. Runa Ghosh Auddy for discussions and English language editing.

References 1) Scannell J. W., Blanckley A., Boldon H., Warrington B., Nat. Rev.

Drug Discov., 11, 191–200 (2012). 2) Anand P., Kunnumakkara A. B., Newman R. A., Aggarwal B. B.,

Mol. Pharm., 4, 807–818 (2007). 3) Ma Y., Zhao X., Li J., Shen Q., Int. J. Nanomedicine, 7, 559–570

(2012). 4) Kamaly N., Xiao Z., Valencia P. M., Radovic-Moreno A. F.,

Farokhzad O. C., Chem. Soc. Rev., 41, 2971–3010 (2012). 5) Huynh N. T., Morille M., Bejaud J., Legras P., Vessieres A., Jaouen

G., Benoit J. P., Passirani C., Pharm. Res., 28, 3189–3198 (2011). 6) Cho K. J., Moon H. T., Park G. E., Jeon O. C., Byun Y., Lee Y. K.,

Bioconjug. Chem., 19, 1346–1351 (2008). 7) Deepagan V. G., Sarmento B., Menon D., Nascimento A., Jayasree

A., Sreeranganathan M., Koyakutty M., Nair S. V., Rangasamy J., Nanomedicine (Lond.), 7, 507–519 (2012).

8) Jain R. A., Biomaterials, 21, 2475–2490 (2000). 9) Hu B., Pan C., Sun Y., Hou Z., Ye H., Zeng X., J. Agric. Food

Chem., 56, 7451–7458 (2008).10) Low Y., Uehara T., Minowa Y., Yamada H., Ohno Y., Urushidani T.,

Sedykh A., Muratov E., Kuz’min V., Fourches D., Zhu H., Rusyn I., Tropsha A., Chem. Res. Toxicol., 24, 1251–1262 (2011).

11) Rizzi A., Fioni A., J. Chem. Inf. Model., 48, 1686–1692 (2008).12) Faller B., Ertl P., Adv. Drug Deliv. Rev., 59, 533–545 (2007).13) Huynh L., Grant J., Leroux J. C., Delmas P., Allen C., Pharm. Res.,

25, 147–157 (2008).14) Costache A. D., Sheihet L., Zaveri K., Knight D. D., Kohn J., Mol.

Pharm., 6, 1620–1627 (2009).15) Fourches D., Pu D., Tropsha A., Comb. Chem. High Throughput

Screen., 14, 217–225 (2011).16) Wang S., Dormidontova E. E., Biomacromolecules, 11, 1785–1795

(2010).17) Le T., Epa V. C., Burden F. R., Winkler D. A., Chem. Rev., 112,

2889–2919 (2012).18) Jain R. A., Biomaterials, 21, 2475–2490 (2000).19) Roy P., Das S., Bera T., Mondol S., Mukherjee A., Int. J. Nanomedi-

cine, 5, 1113–1121 (2010).20) Shibano M., Lin A. S., Itokawa H., Lee K. H., J. Nat. Prod., 70,

1424–1428 (2007).21) Novotarskyi S., Sushko I., Körner R., Pandey A. K., Tetko I. V., J.

Chem. Inf. Model., 51, 1271–1280 (2011).22) Viswanadhan V. N., Ghose A. K., Revankar G. R., Robins R. K., J.

Chem. Inf. Comput. Sci., 29, 163–172 (1989).23) Balaban A. T., Ciubotariu D., Medeleanu M., J. Chem. Inf. Comput.

Sci., 31, 517–523 (1991).24) Hemmer M. C., Steinhauer V., Gasteiger J., Vib. Spectrosc., 19,

151–164 (1999).25) Schuur J. H., Selzer P., Gasteiger J., J. Chem. Inf. Comput. Sci., 36,

334–344 (1996).26) Todeschini R., Lasagni M., Marengo E., J. Chemom., 8, 263–273

(1994).27) Consonni V., Todeschini R., Pavan M., J. Chem. Inf. Comput. Sci.,

42, 682–692 (2002).28) Randic´ M., J. Am. Chem. Soc., 97, 6609–6615 (1975).29) Kier L. B., Quant. Struct.-Act. Relat., 4, 109–116 (1985). 30) Kier L. B., Quant. Struct.-Act. Relat., 6, 8–12 (1987). 31) Petitjean M., J. Chem. Inf. Comput. Sci., 32, 331–337 (1992).

32) Wiener H., J. Am. Chem. Soc., 69, 17–20 (1947).33) Wiener H., J. Am. Chem. Soc., 69, 2636–2638 (1947).34) Balaban A. T., Theoret. Chim. Acta, 53, 355–375 (1979).35) Balaban A. T., Chem. Phys. Lett., 89, 399–404 (1982).36) Gasteiger J., Marsili M., Tetrahedron, 36, 3219–3228 (1980).37) Stanton D. T., Jurs P. C., Anal. Chem., 62, 2323–2329 (1990).38) Cruciani G., Crivori P., Carrupt P. A., Testa B., J. Mol. Struc.

Theochem., 503, 17–30 (2000).39) Wold S., Eriksson L., “Chemometric Methods in Molecular De-

sign,” Vol. II, ed. by de Waterbeemd H. V., VCH Verlagsgesell-schaft mbh, Weinheim, 1995, p. 312.

40) Catalyst, Version 4.11, Accelrys Inc., San Diego, CA, 2005. ‹http://www.accelrys.com›.

41) Ibrahim H. K., El-Leithy I. S., Makky A. A., Mol. Pharm., 7, 576–585 (2010).

42) Nagar S., Saha A., Eur. J. Med. Chem., 45, 4307–4315 (2010).43) Budhian A., Siegel S. J., Winey K. I., Int. J. Pharm., 336, 367–375

(2007).44) Segundo P. E., Quintanar A. G., Perez V. A., Guerrero D. Q., Int. J.

Pharm., 294, 217–232 (2005). 45) Nah J. W., Jeong Y. I., Koh J. J., Korean J. Chem. Eng., 17, 230–236

(2000).46) Moraes C. M., de Matos A. P., de Lima R., Rosa A. H., de Paula E.,

Fraceto L. F., J. Biol. Phys., 33, 455–461 (2007).47) Mu L., Feng S. S., Pharm. Res., 20, 1864–1872 (2003).48) Dhar S., Gu F. X., Langer R., Farokhzad O. C., Lippard S. J., Proc.

Natl. Acad. Sci. U.S.A., 105, 17356–17361 (2008). 49) Muthu M. S., Rawat M. K., Mishra A., Singh S., Nanomedicine, 5,

323–333 (2009).50) Lamprecht A., Yamamoto H., Takeuchi H., Kawashima Y., J. Con-

trolled Release, 90, 313–322 (2003).51) Song X., Zhao Y., Hou S., Xu F., Zhao R., He J., Cai Z., Li Y., Chen

Q., Eur. J. Pharm. Biopharm., 69, 445–453 (2008).52) Teixeira M., Alonso M. J., Pinto M. M. M., Barbosa C. M., Eur. J.

Pharm. Biopharm., 59, 491–500 (2005).53) Ameller T., Marsaud V., Legrand P., Gref R., Barratt G., Renoir J.

M., Pharm. Res., 20, 1063–1070 (2003).54) Duan Y., Sun X., Gong T., Wang Q., Zhang Z., J. Mater. Sci. Mater.

Med., 17, 509–516 (2006).55) Govender T., Stolnik S., Garnett M. C., Illum L., Davis S. S., J.

Controlled Release, 57, 171–185 (1999).56) Mondal N., Halder K. K., Kamila M. M., Debnath M. C., Pal T.

K., Ghosal S. K., Sarkar B. R., Ganguly S., Int. J. Pharm., 397, 194–200 (2010).

57) Joshi S. A., Chavhan S. S., Sawant K. K., Eur. J. Pharm. Bio-pharm., 76, 189–199 (2010).

58) Gupta H., Aqil M., Khar R. K., Ali A., Bhatnagar A., Mittal G., Nanomedicine, 6, 324–333 (2010).

59) Yallapu M. M., Gupta B. K., Jaggi M., Chauhan S. C., J. Colloid Interface Sci., 351, 19–29 (2010).

60) Nagar S., Saha A., J. Comput. Chem., 31, 2342–2353 (2010).61) Islam M. A., Nagar S., Das S., Mukherjee A., Saha A., Biol. Pharm.

Bull., 31, 1453–1460 (2008).62) Todeschni R., Consonni V., “Handbook of Molecular Descriptors,”

Vol. XI, ed. by Mannhold R., Kubinyi H., Timmerman H., Wiley-VCH, Weinheim, 2000, p. 667.

63) Yang L., Alexandridis P., Curr. Opin. Colloid Interface Sci., 5, 132–143 (2000).

64) Jain G. K., Pathan S. A., Akhter S., Ahmad N., Jain N., Talegaonkar S., Khar R. K., Ahmad F. J., Polym. Degrad. Stabil., 95, 2360–2366 (2010).

65) Sant S., Thommes M., Hildgen P., Langmuir, 24, 280–287 (2008).66) Korsmeyer R. W., Gurny R., Doelker E. M., Buri P., Peppas N. A.,

Int. J. Pharm., 15, 25–35 (1983). 67) Grattard N., Pernin M., Marty B., Roudaut G., Champion D., Le

Meste M., J. Controlled Release, 84, 125–135 (2002).