Oo

NWa

b

c

d

e

a

ARRAA

KTCAXS

1

tdmcttch

id

h0

International Journal of Pharmaceutics 494 (2015) 296–303

Contents lists available at ScienceDirect

International Journal of Pharmaceutics

j ourna l h om epa ge: www.elsev ier .com/ locate / i jpharm

ptimization of pH-independent chronotherapeutic releasef verapamil HCl from three-layer matrix tablets

izar Al-Zoubia,∗, Hatim S. Alkhatibb, Ghadah Alobaidi c, Safwan Abdel-Rahimc,asfy Obeidatd, Stavros Malamatarise

Faculty of Pharmaceutical Sciences, Hashemite University, Zarqa, JordanDepartment of Pharmaceutics and Pharmaceutical Technology, Faculty of Pharmacy, The University of Jordan, Amman, JordanDepartment of Pharmaceutical Sciences and Pharmaceutics, Faculty of Pharmacy, Applied Science University, Amman, JordanDepartment of Pharmaceutical Technology, Faculty of Pharmacy, Jordan University of Science and Technology, Irbid, JordanDepartment of Pharmaceutical Technology, School of Pharmacy, Aristotle University of Thessaloniki, Thessaloniki, Greece

r t i c l e i n f o

rticle history:eceived 2 March 2015eceived in revised form 1 August 2015ccepted 8 August 2015vailable online 11 August 2015

eywords:hree-layer matrix tabletshronodeliveryrtificial neural networksanthan gumodium alginate

a b s t r a c t

The aim of this work was to evaluate and optimize formulation of three-layer matrix tablets based onxanthan gum (XG) and sodium alginate for chronotherapeutic pH-independent release of verapamil HCl(VH). Artificial neural networks (ANN) were applied in the optimization and compared with multiplelinear regression (MLR). A face-centered central composite experimental design was employed with threefactors (mass fraction of VH in intermediate layer, X1, and of XG in matrix former of intermediate andouter layers, X2 and X3). The prepared tablets were tested for in vitro release in 0.1 N HCl and phosphatebuffer (pH 7.5), tensile strength and friability. Furthermore, swelling observation and release modelingto Weibull function and power law equation of Peppas were employed to help further understanding ofrelease behavior and mechanism. The releases (%) in phosphate buffer (pH 7.5) at 6, 12 and 24 h wereselected as responses to depict the mode of release and similarity factor (f2), between release profilesin 0.1 N HCl and pH 7.5 during the first 8 h, as response of pH-independence. A desirability function

combining the four responses was constructed and overall desirability values were used for the ANN andMLR modeling. Five additional checkpoint formulations, within the experimental domain, were used tovalidate the external predictability of the models. The constructed ANN model fitted better to the overalldesirability than the MLR model (R = 0.838 vs. 0.670, for the additional checkpoint formulations) andtherefore, was used for prediction of formulation with optimal in vitro drug release.

. Introduction

Blood pressure (BP) is characterized by regular changes duringhe day (circadian rhythm), where it increases at morning, whilerops down during nocturnal sleep. This circadian rhythm of BPakes the relatively constant drug level in plasma, achieved by

onventional controlled-release dosage forms, not ideal for thereatment of hypertension. Chronotherapeutic drug delivery sys-ems which release the drug at a slow rate during the night sleepan control plasma concentrations of antihypertensive drugs inarmony with the BP circadian rhythm (Hermida et al., 2007).

Verapamil hydrochloride (VH) is a calcium channel blocker usedn the treatment of cardiac arrhythmia, angina pectoris, myocar-ial infarction and hypertension as well. It is a weakly-basic drug

∗ Corresponding author. Tel.: +962 5 3903333; fax: +962 5 3903368.E-mail address: nzoubi@hu.edu.jo (N. Al-Zoubi).

ttp://dx.doi.org/10.1016/j.ijpharm.2015.08.021378-5173/© 2015 Elsevier B.V. All rights reserved.

© 2015 Elsevier B.V. All rights reserved.

characterized by pH-dependent solubility, which was encounteredby several formulation approaches: (i) incorporation of weak acidsin matrix formulations as pH modifiers (Streubel et al., 2000a);(ii) development of floating systems extending the gastric empting(Soppimath et al., 2001; Streubel et al., 2002, 2003); (iii) combina-tion of enteric and release extending polymers applied as coatingon pellets for membrane-controlled release (Dashevsky et al., 2004)or incorporated in different matrix system (Streubel et al., 2000a;Tatavarti et al., 2004); and (iv) development of hydrophilic matriceswith pH-dependent swelling based on sodium alginate (Al-Zoubiet al., 2011; Gutsche et al., 2008; Timmins et al., 1997).

Currently commercial VH modified release products are avail-able either as matrix tablets based on sodium alginate (Isoptin®

SR and Calan® SR) or as reservoir multiparticulates in capsules

(Verelan® and Verelan® PM) and osmotic pump systems (Covera-HS®). Of these commercial products Verelan® PM and Covera-HS®

are chronotherapeutic systems based on CODASTM and OROS®

Push-PullTM technologies, respectively (Qiu and Zhang, 2009).

N. Al-Zoubi et al. / International Journal of Pharmaceutics 494 (2015) 296–303 297

Table 1Composition of three layer matrix-tablets comprising the face-center cube experimental design (indicated by the levels of formulation factors) together with their tensilestrength, friability and kinetic model (power law model and Weibull function) fitting parameters of the release data in phosphate buffer (pH 7.5).

Formulation factors Tensile strength (MPa)a Friability (%) Power law parameters Weibull parameters

X1 X2 X3 n Kp r b td r

1 1.0 1.0 1.0 2.57 0.11 0.161 0.105 0.906 0.175 291990.8 0.9032 1.0 1.0 0.0 1.05 0.65 0.663 0.101 0.922 0.993 10.3 0.9703 1.0 0.0 1.0 1.68 0.29 0.778 0.012 0.970 0.476 1281.7 0.9604 1.0 0.0 0.0 0.82 1.00 1.204 0.058 0.993 1.428 10.0 0.9905 0.2 1.0 1.0 1.18 0.47 0.261 0.283 0.922 0.337 25.5 0.9396 0.2 1.0 0.0 1.01 0.50 0.592 0.101 0.967 0.685 26.5 0.9767 0.2 0.0 1.0 0.85 0.34 0.478 0.217 0.999 0.680 8.2 0.9948 0.2 0.0 0.0 0.60 0.41 1.106 0.103 0.996 1.132 6.4 0.9859 0.6 0.5 0.5 1.21 0.31 0.962 0.071 0.987 1.085 11.1 0.99410 0.6 1.0 0.5 1.21 0.63 0.658 0.088 0.936 1.036 11.6 0.96211 0.6 0.0 0.5 0.94 0.75 1.004 0.100 0.990 1.056 7.9 0.99212 1.0 0.5 0.5 1.27 0.50 1.054 0.032 0.990 1.248 16.5 0.98913 0.2 0.5 0.5 1.02 0.37 1.062 0.096 0.996 1.236 6.4 0.998

nBtc(1(otoiamftlftdmto

2

2

aatau

2

tota(pS

14 0.6 0.5 1.0 1.07 0.25

15 0.6 0.5 0.0 0.80 1.31

a Standard deviation <0.2.

Multi-layered matrix tablets are a recognized flexible tech-ology for modifying release of drugs (Abdul and Poddar, 2004;aloglu and S enyigit, 2010; Efentakis and Peponaki, 2008) and par-icularly for chronotherapeutic delivery (Khan et al., 2013). Theyan result to different modes of release, e.g. to zero-order releaseChidambaram et al., 1998; Conte et al., 1992a, 1992b; Qiu et al.,998), to pulsatile (Conte and Maggi, 2000) or to bimodal releaseStreubel et al., 2000b). Recently, three-layer matrix tablets basedn mixtures of xanthan gum (XG) and sodium alginate (SA) con-aining altered drug content were proposed as a simple approachf release programing for a model drug (diltiazem HCl) of pH-ndependent solubility (Al-Zoubi and Malamataris, 2008). Such

system, combining three-layer tablet technology and polymerixture of pH-dependent swelling, was thought to be of interest

or application in chronotherapeutic oral delivery of VH charac-erized by pH-dependent solubility, since it seems simpler andess expensive in comparison with the currently marketed dosageorms based on reservoir and osmotic pump design. Therefore, inhe present work a face-centered central composite experimentalesign is applied in order to evaluate and optimize a three-layeratrix-tablet formulation based on sodium alginate (SA) and xan-

han gum (XG) for pH-independent, chronotherapeutic releasef VH.

. Materials and methods

.1. Materials

Verapamil HCl powder purchased from Medex, UK, was used asctive pharmaceutical ingredient (API). Xanthan gum and sodiumlginate powders purchased from Sigma, USA, and BDH, UK, respec-ively, were used as the hydrophilic matrix former components. Forll the materials, size fraction <180 �m was obtained by sieving andsed for the preparation of the three-layer matrix tablets.

.2. Preparation of verapamil HCl three-layer matrix tablets

Fifteen different experimental formulae of VH three-layerablets comprising a 400 mg middle layer and two 200-mguter layers of similar composition were prepared accordingo a face-centered cube design described previously (Al-Zoubi

nd Malamataris, 2008). All the tablets contained fixed amount300 mg) of VH distributed in the middle and the outer layers afterhysical mixing by spatulation for 15 min with matrix former (XG,A or binary mixtures of them). A manual hydraulic press (Reiken

1.036 0.030 0.993 1.213 18.7 0.9911.090 0.081 0.996 1.131 8.3 0.987

Seiki, Japan) and 13-mm round flat-faced punch and die set wereused for compression. The first two layers (outer and intermedi-ate) were uniformly spread one after another in the die by slightlycompressing each layer at 2.8 kN for 5 s. Then, the third layer (outer)was added and tablets were compressed at a force of 14 kN for 30 s.Three formulation factors at three levels were applied: the massfraction of VH incorporated in the intermediate layer, X1, (0.2, 0.6and 1) and the mass fraction of XG in the matrix former of inter-mediate and of outer layers, X2 and X3, (0, 0.5 and 1.0) shown inTable 1.

2.3. In vitro drug release

The release study was performed according to USP, in a pad-dle dissolution system (Apparatus II), at 50 rpm, using 900 ml ofdissolution medium (0.1 N HCl or phosphate buffer, pH 7.5) at atemperature of 37 ± 0.5 ◦C. At predetermined time intervals, sam-ples were withdrawn with replacement of the dissolution medium,filtered and the concentration of dissolved VH was assessed aftersuitable dilution, by UV spectroscopy at a wavelength correspond-ing to absorbance maximum (278 nm). All tests were performed intriplicate and from the mean concentration, the drug release (%)was determined.

The cumulative Weibull distribution function was fitted to therelease results with the aid of MS Excel employing the linearizedform (Bonferoni et al., 1998; Langenbucher, 1972):

log [− ln (1 − m)] = b log (t − Ti) − log a (1)

where m is the cumulative drug release at time t, Ti is the lag timebefore the onset of dissolution or release process that in most caseswill be zero, b is the shape parameter and a is the time scale of theprocess. From a and b the time for 63.2% release, td, can be calculated(Costa and Souza Lobo, 2001):

td = a1/b (2)

The Weibull function was selected in order to characterize theshape of the release profile, since it is capable of dealing with dis-solution profiles corresponding to an initial release phase followedby either a faster or slower one. More specifically, when b = 1, the

curve is exponential and Eq. (1) reduces to the simple first ordermodel. For b > 1, the release profile is sigmoid, while for b < 1 it isparabolic corresponding to fast initial release slowing gradually. Forcharacterization of the release mechanism, the power law model

2 nal of

od

widFdz(

2

fhi

T

woc

Padai

%

2

wiseasio

2m

rtmma

f

war

ri

98 N. Al-Zoubi et al. / International Jour

f Peppas was fitted by non-linear regression using data up to 60%rug release (Peppas, 1985):

Mt

M∞= Kp · tn (3)

here Mt/M∞ represents the fractional release of drug at time t, Kp

s the release rate constant and n is the release exponent. For cylin-rical tablets, like those in this study, a value of n ≤ 0.45 indicatesickian diffusion, 0.45 < n < 0.89 indicates non-Fickian (anomalous)iffusion, n = 0.89 indicates Case-II transport (erosion control andero order kinetics) and n > 0.89 indicates Super Case-II transportRitger and Peppas, 1987).

.4. Tensile strength and friability

Tensile strength, Ts, of the three-layer tablets was determinedrom the force applied for diametral breaking in an Erweka TBH 325ardness tester (Erweka GmbH, Heusenstamm, Germany) accord-

ng to Fell and Newton:

s = 2F

�DT(4)

here F is the breaking force, D and T are the diameter and thicknessf the tablets, respectively (Fell and Newton, 1970). Mean value wasalculated from five replicas.

Friability was measured using a tablet friability tester (PTF 20E,harmatest, Germany). Ten tablets were carefully dedusted using

fine brush, accurately weighed, and placed in the drum. Then, therum was rotated 100 times. The tablets were dedusted again andccurately weighed. The friability percentage was calculated fromnitial and final weights:

Friability =(

weight before − weight afterweight before

)∗ 100 (5)

.5. Observation of swelling

Changes of tablets during dissolution testing due to swellingere observed and photographs were taken after 1 h immersion

n the dissolution media. Tablets were gently withdrawn using apoon spatula and a filter paper was used to remove carefully thexcess liquid from around the swollen tablets avoiding touchingnd deforming the jelly layer. Photographs were taken (top andide views) using a digital camera (EasyShare Z1285, KODAK, USA)n order to observe changes in diameter and thickness or integrityf layers.

.6. Responses (dependent variables) and desirability function forulti-objective optimization

The results of % release in buffer, at 6, 12 and 24 h (Y1 to Y3,espectively) were selected as responses for the optimization. Fur-hermore the similarity of the release profile in acidic and buffer

edia for the first 8 h determined as the factor (f2) was selected aseasure of pH-independence of VH release, Y4. f2 was calculated

ccording to the following equation (Moore and Flanner, 1996):

2 = 50 log

⎧⎨⎩

[1 + 1

n

n∑t=1

(Rt − Tt)2

]−0.5

× 100

⎫⎬⎭ (6)

here n is the number of dissolution time points and Rt and Tt

re the dissolution values, at time t, in acidic and buffered media,

espectively.

Multi-objective optimization was sought for an initially slowelease phase (with minimum pH-dependence) followed by anncreased release rate ending with an almost complete drug

Pharmaceutics 494 (2015) 296–303

release at 24 h. For this purpose, individual desirability functions(d1–d4) of Smaller-The-Best (for Y1), Nominal-The-Best (for Y2) andLarger-The-Best (for Y3) and for pH-independence (f2 or Y4) wereestablished based on the following equations (Derringer and Suich,1980):

d1 = (U − Y1)(U − L)

(7)

d2 =

⎧⎪⎪⎪⎪⎨⎪⎪⎪⎪⎩

(Y2 − L)(U − L)

if Y2 < 55

1 if 55 < Y2 < 65

(U − Y1)(U − L)

if Y2 < 65

(8)

d3 = (Y3 − L)(U − L)

(9)

d4 = (Y4 − L)(U − L)

(10)

where U and L are the upper and lower specification limits ofthe response, respectively. U and L values for each function wereselected so that the desirability values for the worse and bestresponse to be close to 0 and 1, respectively, and calculated usingthe following equations:

U = Ymax + 0.05 × (Ymax − Ymin) (11)

L = Ymin − 0.05 × (Ymax − Ymin) (12)

where Ymax and Ymin are the maximum and minimum responsevalues observed for the fifteen experimental formulae (Bodea andLeucuta, 1998).

For the optimization, an overall desirability function (Y5) wasused combining all the individual desirability functions of theselected responses to one value, Y5, (Derringer, 1994; Derringerand Suich, 1980):

Y5 = (d1 × d2 × d3 × d4)1/4 (13)

2.7. Computational methods for modelling of overall desirability

Two optimization techniques were evaluated regarding therelation between the overall desirability values and the indepen-dent experimental factors: A generalized feed forward multilayerperception artificial neural network (ANN) and multiple linearregression (MLR) of second-order polynomial equation by usingthe software Neurosolutions® (Neurodimensions Inc., USA) and theSPSS 22.0 program (IBM Inc. Chicago, IL, USA), respectively. The pro-posed ANN and MLR models were validated by employing five testpoints. Three-layer tablets of composition corresponding to thesefive test points were prepared and evaluated similarly to the 15formulae used in modeling. Subsequently, the experimental overalldesirability values based on the selected responses were comparedwith model-predicted values.

3. Results and discussion

3.1. Release and physicomechanical properties of three layermatrix-tablets

The release profiles for the 15 experimental formulas compris-ing the face centered cubic design are shown in Fig. 1, divided in

three groups according to the mass fraction of XG incorporatedin the outer layers, X3, (1, 0.5 and 0). The results of mechanicaltesting (tensile strength and friability) and kinetic release model-ing for the fifteen experimental tablets (F1–F15) are summarized

N. Al-Zoubi et al. / International Journal of Pharmaceutics 494 (2015) 296–303 299

F tereda 1). Er

ienisfca

baes

ig. 1. Release profiles for the 15 experimental formulae comprising the face cenccording to the mass fraction of XG incorporated in the outer layers (X3, 0, 0.5 and

n Table 1, together with the formulation factors and their lev-ls applied. The results of the parameters used in optimization,amely release parameters (Y1–Y3), the release similarity or pH-

ndependence (Y4), and the overall desirability function (Y5) areummarized in Table 2. Photographs of top and side views of tabletormulations corresponding to the corner points of the cubical face-entered experimental design (F1–F8), after 1 h of immersion incidic and buffered media, are shown in Fig. 2.

All formulas had sufficient mechanical strength as indicated

y their low friability values. Regarding crushing resistance test,ll formulas showed diametrical breakage and no splitting of lay-rs was observed indicating good adhesion of layers. The tensiletrength was generally increased by increasing mass fraction of VH

cubic design in 0.1 N HCl and phosphate buffer media (full and empty symbols),ror bars represent standard deviation (n = 3).

in the intermediate layer and mass fraction of XG in the outer andintermediate layers.

From the release profiles, Fig. 1, and the release parameters tdand n in Table 1 and Y1 to Y3 in Table 2, it can be seen that theincrease of XG mass fraction in the outer layers is resulting in reduc-tion of the release rate and in shifting of release mechanism towardFickian diffusion for most of the formulations. This is probably dueto formation of stronger and less erodible gel barrier and is in con-sistence with previous findings (Al-Zoubi and Malamataris, 2008).

Furthermore, Fig. 1 and Tables 1 and 2 show that release is reducedwith the increase of VH mass fraction in the intermediate layer,X1. More specifically, the release in buffer was over-slow (releaseafter 24 h Y3 < 40%, Table 2) for formulae 1 and 3 (with all VH in the

300 N. Al-Zoubi et al. / International Journal of Pharmaceutics 494 (2015) 296–303

F e corb

i56rfadr

Vi

ig. 2. Top and side views of VH three-layer tablet formulations corresponding to thuffered media.

ntermediate layer and only XG in the outer layer) and for formula (with only XG as matrix former). On the other hand, formulae 2, 4,, 8 and 15 (containing no XG in the outer layers) showed an over-apid release in buffer (release after 12 h Y2 > 80%, Table 1). Alsoormulae with 1:1 XG:SA blend as matrix former of the outer layersnd low or intermediate level of VH mass fraction in the interme-iate layer (formulae 11 and 13, respectively) exhibited over-rapid

elease.

Due to the known pH-dependent solubility of the weakly-basicH (Streubel et al., 2000a), a higher release in acid than in buffer

s expected for the hydrophilic matrix systems investigated unless

ner points of face-centered cube design (F1–F8) after 1 h of immersion in acidic and

that the swelling and erosion behavior of matrix former(s) is capa-ble to compensate for the increased solubility. By consideringhigher than 50 f2 value as indication of release similarity betweenthe dissolution data in acidic and buffer media (pH-independence),the results in Table 2 show that only for six formulae the f2 valueis lower than 50. More specifically, the lowest f2 values were foundfor tablets with intermediate layer containing all the amount of VH

dispersed in SA only as matrix former (f2 = 38.0 and 19.9 for formu-lae 3 and 4, respectively, Table 2), which showed higher release inacid than in buffer. This can be attributed to rapid dissolution of VHthroughout the sides of intermediate layer facilitated by the high

N. Al-Zoubi et al. / International Journal of

Table 2Selected responses (parameters of: % release, Y1 to Y3, pH-independence of release,Y4, and overall desirability, Y5) for optimized chronotherapeutic delivery.

Formulation factors Responses

X1 X2 X3 Y1 Y2 Y3 Y4 Y5

(%) (%) (%) (f2)

1 1.0 1.0 1.0 13.3 16.3 22.8 72.2 0.2792 1.0 1.0 0.0 48.8 81.0 83.8 56.8 0.5973 1.0 0.0 1.0 6.2 11.1 34.9 38.0 0.2354 1.0 0.0 0.0 56.9 96.5 98.3 19.9 0.1685 0.2 1.0 1.0 11.1 14.5 20.9 40.1 0.1756 0.2 1.0 0.0 68.9 85.9 91.7 51.0 0.4597 0.2 0.0 1.0 52.4 78.1 86.1 47.0 0.5708 0.2 0.0 0.0 86.3 93.6 94.3 43.2 0.2239 0.6 0.5 0.5 32.8 72.1 85.8 73.4 0.79010 0.6 1.0 0.5 26.9 56.9 92.5 59.0 0.81811 0.6 0.0 0.5 51.0 84.6 91.5 56.2 0.56812 1.0 0.5 0.5 17.8 54.6 79.0 74.7 0.86913 0.2 0.5 0.5 58.4 90.7 100.7 44.2 0.435

dgav2tsiefotabt

cai1maaacXFSraac

values were in good agreement with the observed experimental

TCd

14 0.6 0.5 1.0 18.2 38.7 86.7 67.4 0.77715 0.6 0.5 0.0 58.7 87.1 90.7 54.2 0.503

rug content (only 25% polymer content), the lack of a controllingel barrier (due to transformation of SA to the much less swellablelginic acid), the formation of cracks (Fig. 2) which has been pre-iously reported (Efentakis and Buckton, 2002; Sriamornsak et al.,007; Al-Zoubi et al., 2011) and the separation (lamination) of thehree layers clearly seen for formula 4, Fig. 2. The other tablets thathowed pH-dependent release (f2 < 50) contained most of the drugn the outer layer (formulae 5, 7, 8 and 13, Table 2). This can bexplained by the elimination of the hydrophilic barrier and there-ore of diffusion for the dissolved VH molecules and predominancef direct VH particle dissolution, which is much more affected byhe VH solubility in the dissolution media known to be higher incidic pH (Streubel et al., 2000a). The elimination of the hydrophilicarriers should be caused by the quicker erosion of outer layers (dueo decreased polymer-to-drug ratio)

Taking into account that the desired release profile has to beharacterized by an initial slow release followed by accelerationfter about 6 h, it requires that all or most of the drug to be includedn the intermediate layer and was most clearly shown by formula2 containing all VH in the intermediate layer together with 1:1ixture of SA and XG as the matrix former. In this case the release

cceleration might be attributed to erosion of the outer layers afterbout 6 h. Therefore we can conclude that a binary mixture of SAnd XG seems necessary for the chronotherapeutic delivery. In theontrary, erosion of outer layers is very slow when they contain onlyG as matrix former (formulae 1 and 3) leading to very low release,ig. 1, and decreased erosion (Fig. 2), while tablets containing onlyA in the outer layers as matrix former exhibit very early and fastelease acceleration (formulae 2 and 4) although they also contain

ll the amount of VH in the intermediate layer. The later should bettributed to rapid erosion of outer layer (Fig. 2) leading to almostomplete release at 12 h.

able 3omposition of the additional five checkpoint formulations (indicated by the levels of

esirability applying ANN and MLR models and corresponding residuals (Y5).

Tablets Formulation factors

X1 X2 X3

Test 1 0.8 0.8 0.4

Test 2 0.4 0.4 0.6

Test 3 0.5 0.4 0.2

Test 4 0.3 0.7 0.3

Test 5 0.8 0.5 0.0

Pharmaceutics 494 (2015) 296–303 301

3.2. Development of models

3.2.1. Artificial neural networks (ANN) modelingThree inputs, corresponding to the examined formulation fac-

tors (independent variables), and one output unit, corresponding tothe overall desirability value, were used in the development of ANNmodel. A generalized feed forward multilayer perception networkwas employed and the hyperbolic tangent function was selected astransfer function. The Levenberg–Marquardt algorithm was usedas learning rule for all layers.

Preliminary trial-and-error testing was used to determinesuitable network structure through step-by-step increase of thenumber of processing elements in the hidden layer (up to 10) andthe number of hidden layers (up to 3). First, the rows were ran-domized and then the first 12 rows were tagged as training andthe last 3 rows were tagged as cross-validation. The network wastrained three times using new random sets of initial weights andeach cycle consisted of 1000 epochs (iterations). The training end-point was selected as the point corresponding to the lowest MeanSquared Error (MSE) of the validation subset to avoid over-training.To select the optimal ANN model, linear trend lines were created forthe observed versus predicted responses for the five test samples,which were not used in the training. The ANN model that yieldeda regression plot with a squared correlation coefficient (R2) thatwas closest to 1.0 was selected as the optimal (Ibric et al., 2007;Djekic et al., 2008). The best network consisted of one hidden layercontaining six processing elements.

3.2.2. Multiple linear regression (MLR) modelRelations between the overall desirability values (Y5) and the

independent experimental variables (formulation factors) weresought by applying multiple linear regression (MLR) and fittingof second-order polynomial equation including two-factor inter-action terms. In order to obtain the best MLR model, backwardelimination procedure was applied to simplify the obtained poly-nomial equation by removing non-significant terms (p > 0.05). Theoptimal model was selected as that of highest adjusted coefficientof determination, R2

adj (lowest standard error of estimates, SEE),R = 0.864 and p = 0.040:

Y5 = 0.051 + 1.344X1 + 1.103X3 − 1.275X21 − 0.864X2

3

+ 0.515X1X2 − 0.444X2X3 (14)

3.2.3. Comparison of ANN and MLR modelsThe composition of the five test points is listed in Table 3

together with the predicted, the experimental, and the residualvalues of the overall desirability (Y5). In general, the predicted

values and the residuals varied from −0.065 to +0.105 and from−0.271 to +0.222 for ANN and MLR, respectively. The correla-tion between predicted and observed values of Y5 was better in

formulation factors) together with experimental and predicted values of overall

Overall desirability values

Experimental Predicted and (Residuals) by

ANN MLR

0.587 0.652 (−0.065) 0.801 (−0.214)0.440 0.472 (−0.032) 0.711 (−0.271)0.551 0.535 (0.016) 0.658 (−0.107)0.745 0.645 (0.100) 0.608 (0.138)0.738 0.633 (0.105) 0.516 (0.222)

302 N. Al-Zoubi et al. / International Journal of Pharmaceutics 494 (2015) 296–303

F nd XGo

ttA

3

g(ate

ig. 3. Effect of the mass fraction of VH incorporated in the intermediate layer (X1) an the overall desirability value (Y5) as predicted by ANN (left) and MLR (right).

he case of ANN than MLR (R = 0.838 and 0.670, respectively, forhe additional checkpoint formulations) indicating superiority ofNN.

.2.4. Multi-objective optimizationFig. 3 presents contour plots showing the effects of the investi-

ated formulation factors (X1–X3) on the overall desirability value

Y5), which are based on the data derived by the proposed ANNnd MLR models. It shows the nonlinear relationship betweenhe formulation factors and the predicted release parameters bymploying the ANN, whereas using the second-order polynomial

in the matrix former of the intermediate and outer layers (X2 and X3, respectively)

equations (MLR model) the plots are characterized by almost par-allel contours.

The optimal tablet formula with maximum overall desirabil-ity value, predicted by the ANN model, Y5 = 0.894, correspondsto X1 = 1.0, X2 = 0.65 and X3 = 0.55; and experimental release pro-files (in acidic and buffered media) of this optimal formula areshown in Fig. 4. For this optimal formula the experimental values

of individual desirability are Y1 = 15.2, Y2 = 50.9, Y3 = 89.3, Y4 = 72.2and the overall desirability (Y5 = 0.873) is slightly above the high-est value obtained with the applied experimental set (Y5 = 0.869)corresponding to formula 12, which contains all the drug in the

N. Al-Zoubi et al. / International Journal of

Fo(

i(ltiom

4

waaimragottd

R

A

A

A

B

B

B

C

C

ig. 4. Experimental in vitro release of VH in acidic and buffered media for theptimal formula defined by the ANN model. Error bars represent standard deviationn = 3).

ntermediate layer and binary physical mixtures of XG and SA0.5:0.5 w/w) as matrix formers of the intermediate and outerayers. This is in agreement with their close composition and fur-hermore the small changes of Y5 at high level of VH mass fractionn the intermediate layer (X1 = 1.0), (left), indicate robustness of theptimal formula to variations in the distribution uniformity of theatrix former components.

. Conclusions

Three-layer matrix tablets of acceptable mechanical strength,ith 0.65:0.35 and 0.55:0.45 w/w binary physical mixtures of XG

nd SA as matrix formers of the intermediate and outer layersnd all the drug in the intermediate layer give optimized pH-ndependent chronotherapeutic delivery of VH. XG and SA alone as

atrix formers of outer layers result either in unacceptably slowate and low extent of drug release (in the case of XG) or in very earlynd fast release acceleration (in the case of SA). The ANN modelives better correlation between predicted and observed valuesf overall desirability value (Y5) than MLR (R = 0.838 vs.0.670, forhe additional checkpoint formulations) indicating superiority forhe optimization of pH-independent chronotherapeutic delivery ofrugs characterized by pH-dependent solubility.

eferences

bdul, S., Poddar, S., 2004. A flexible technology for modified release of drugs: multilayered tablets. J. Controlled Release 97, 393–405.

l-Zoubi, N., AlKhatib, H., Obeidat, W., 2011. Evaluation of hydrophilic matrix tabletsbased on Carbopol® 971P and low-viscosity sodium alginate for pH-independentcontrolled drug release. Drug Dev. Ind. Pharm. 37, 798–808.

l-Zoubi, N., Malamataris, S., 2008. Three-layer matrix tablets and simple approachof drug release programming. J. Drug Delivery Sci. Technol. 18, 431–437.

aloglu, E., S enyigit, T., 2010. A design and evaluation of layered matrix tablet for-mulations of metoprolol tartrate. AAPS PharmSciTech 11 (2), 563–573.

odea, A., Leucuta, S.E., 1998. Optimization of propranolol hydrochloride sustained-release pellets using Box–Behnken design and desirability function. Drug Dev.Ind. Pharm. 24, 145–155.

onferoni, M., Rossi, S., Ferrari, F., Bertoni, M., Bolhuis, G., Caramella, C., 1998. Onthe employment of � carrageenan in a matrix system III. Optimization of a �carrageenan–HPMC hydrophilic matrix. J. Controlled Release 51, 231–239.

hidambaram, N., Porter, W., Flood, K., Qiu, Y., 1998. Formulation and characteri-zation of new layered diffusional matrices for zero-order sustained-release. J.

Controlled Release 52, 149–158.

onte, U., Colombo, P., Maggi, L., La Manna, A., 1992a. Compressed barrier layersfor constant drug release in swellable matrix tablets. In: Proceedings of SixthInternational Conference on Pharmaceutical Technology, June 2–4, 1992, Paris,France, Vol. II, pp. 102–111.

Pharmaceutics 494 (2015) 296–303 303

Conte, U., Maggi, L., Giunchedi, P., Sangalli, M.E., La Manna, A., 1992b. Multi-layerhydrophilic matrix tablets as controlled delivery systems. In: Proceedings ofthe Second European Symposium on Controlled Drug Delivery, April 1–3, 1992,Noordwijk, Aan Zee, The Netherlands, pp. 117–120.

Conte, U., Maggi, L., 2000. A flexible technology for the linear, pulsatile and delayedrelease of drugs, allowing for easy accommodation of difficult in vitro targets. J.Controlled Release 64, 263–268.

Costa, P., Souza Lobo, J.M., 2001. Modeling and comparison of dissolution profiles.Eur. J. Pharm. Sci. 13, 123–133.

Dashevsky, A., Kolter, K., Bodmeier, R., 2004. pH-independent release of a basic drugfrom pellets coated with the extended release polymer dispersion Kollicoat® SR30 D and the enteric polymer dispersion Kollicoat® MAE 30 DP. Eur. J. Pharm.Biopharm. 58, 45–49.

Derringer, G., 1994. A balancing act: optimizing a product’s properties. Qual. Prog.27, 51–58.

Derringer, G., Suich, R., 1980. Simultaneous optimization of several response vari-ables. J. Qual. Technol. 12, 214–218.

Djekic, L., Ibric, S., Primorac, M., 2008. The application of artificial neural networksin the prediction of microemulsion phase boundaries in PEG-8 caprylic/capricglycerides based systems. Int. J. Pharm. 361, 41–46.

Efentakis, M., Buckton, G., 2002. The effect of erosion and swelling on the dissolutionof theophylline from low and high viscosity sodium alginate matrices. Pharm.Dev. Technol. 7, 69–77.

Efentakis, M., Peponaki, C., 2008. Formulation study and evaluation of matrix andthree-layer tablet sustained drug delivery systems based on Carbopols withisosorbite mononitrate. AAPS PharmSciTech 9 (3), 917–923.

Fell, J.T., Newton, J.M., 1970. Determination of tablet strength by the diametral com-pression test. J. Pharm. Sci. 59, 688–691.

Gutsche, S., Krause, M., Kranz, H., 2008. Strategies to overcome pH-dependent solu-bility of weakly basic drugs by using different types of alginates. Drug Dev. Ind.Pharm. 34, 1277–1284.

Hermida, R.C., Ayala, D.E., Calvo, C., Portaluppi, F., Smolensky, M.H., 2007.Chronotherapy of hypertension: administration-time-dependent effects oftreatment on the circadian pattern of blood pressure. Adv. Drug Delivery Rev.59, 923–939.

Ibric, S., Jovanovic, M., Djuric, Z., Parojcic, J., Solomun, L., Lucic, B., 2007. Generalizedregression neural networks in prediction of drug stability. J. Pharm. Pharmacol.59, 745–750.

Khan, Z., Choonara, Y.E., Kumar, P., du Toit, L.C., Ndesendo, V.M.K., Pillay, V., 2013.A novel multilayered multidisk oral tablet for chronotherapeutic drug delivery.BioMed Res. Int. 2013, 1–16.

Langenbucher, F., 1972. Linerarization of dissolution rate curves by the Weibulldistribution. J. Pharm. Pharmacol. 24, 979–981.

Moore, J., Flanner, H., 1996. Mathematical comparison of curves with an emphasison in-vitro dissolution profiles. Pharm. Technol. 20, 64–74.

Peppas, N.A., 1985. Analysis of Fickian and non-Fickian drug release from polymers.Pharm. Acta Helv. 60, 110–111.

Qiu, Y., Chidambaram, N., Flood, K., 1998. Design and evaluation of layered diffusionmatrices for zero-order sustained-release. J. Controlled Release 51, 123–130.

Qiu, Y., Zhang, G., 2009. Development of modified-release solid oral dosage forms.In: Qiu, Y., Chen, Y., Zhang, G.G., Liu, L., Porter, W. (Eds.), Developing Solid OralDosage forms: Pharmaceutical Theory and Practice. Academic Press, New York,pp. 501–517.

Ritger, P.L., Peppas, N.A., 1987. A simple equation for description of solute release,II. Fickian and anomalous release from swellable devices. J. Controlled Release5, 37–42.

Soppimath, K.S., Kulkarni, A.R., Aminabhavi, T.M., 2001. Development of hollowmicrospheres as floating controlled-release systems for cardiovascular drugs:preparation and release characteristics. Drug Dev. Ind. Pharm. 27, 507–515.

Sriamornsak, P., Thirawong, N., Korkerd, K., 2007. Swelling, erosion and releasebehavior of alginate-based matrix tablets. Eur. J. Pharm. Biopharm. 66, 435–450.

Streubel, A., Siepmann, J., Bodmeier, R., 2002. Floating microparticles based on lowdensity foam powder. Int. J. Pharm. 241, 279–292.

Streubel, A., Siepmann, J., Bodmeier, R., 2003. Floating matrix tablets based on lowdensity foam powder: effects of formulation and processing parameters on drugrelease. Eur. J. Pharm. Sci. 18, 37–45.

Streubel, A., Siepmann, J., Dashevsky, A., Bodmeier, R., 2000a. pH-independentrelease of a weakly basic drug from water-insoluble and -soluble matrix tablets.J. Controlled Release 67, 101–110.

Streubel, A., Siepmann, J., Peppas, N.A., Bodmeier, R., 2000b. Bimodal drug releaseachieved with multi-layer matrix tablets: transport mechanisms and devicedesign. J. Controlled Release 69, 455–468.

Tatavarti, A.S., Mehta, K.A., Augsburger, L.L., Hoag, S.W., 2004. Influence ofmethacrylic and acrylic acid polymers on the release performance of weakly

basic drugs from sustained release hydrophilic matrices. J. Pharm. Sci. 93,2319–2331.

Timmins, P., Delargy, A.M., Howard, J.R., 1997. Optimization and characterizationof a pH-independent extended-release hydrophilic matrix tablet. Pharm. Dev.Technol. 2, 25–31.