Ultrasound assisted extraction of phenolic compounds from P. lentiscus L. leaves: Comparative study of artificial neural network (ANN) versus degree of experiment for prediction ability of phenolic compounds recovery

(1)

See discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/281645154

Ultrasound assisted extraction of phenolic

compounds from P. lentiscus L. leaves:

Comparative study of artificial neural network

(ANN) versus degree of experiment for

prediction ab...

ARTICLE

in

INDUSTRIAL CROPS AND PRODUCTS · DECEMBER 2015

Impact Factor: 2.84 · DOI: 10.1016/j.indcrop.2015.08.062 READS

155 13 AUTHORS

, INCLUDING:

Dahmoune Farid

Université de Béjaïa

37

PUBLICATIONS

68

CITATIONS

SEE PROFILE

Hocine Remini

Université de Béjaïa

37

PUBLICATIONS

49

CITATIONS

SEE PROFILE

Aoun Omar

Université de Béjaïa

18

PUBLICATIONS

25

CITATIONS

SEE PROFILE

Khodir Madani

Université de Béjaïa

118

PUBLICATIONS

576

CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately.

Available from: Hocine Remini Retrieved on: 31 January 2016

(2)

Contents lists available atScienceDirect

Industrial

Crops

and

Products

j o u r n a l h o m e p a g e :w w w . e l s e v i e r . c o m / l o c a t e / i n d c r o p

Ultrasound

assisted

extraction

of

phenolic

compounds

from

P. lentiscus

L. leaves:

Comparative

study

of

artiﬁcial

neural

network

(ANN)

versus

degree

of

experiment

for

prediction

ability

of

phenolic

compounds

recovery

Farid

Dahmoune

a,b,∗

_,

_Hocine

_Remini

a

_,

_Soﬁane

_Dairi

a

_,

_Omar

_Aoun

a,c

_,

_Kamal

_Moussi

a

_,

Nadia

Bouaoudia-Madi

a

_,

_Nawel

_Adjeroud

a

_,

_Nabil

_Kadri

b

_,

_Khalef

_Lefsih

a

_,

_Lhadi

Boughani

a,b

_,

_Lotﬁ

_Mouni

d

_,

_Balunkeswar

_Nayak

e

_,

_Khodir

_Madani

a

a_Laboratoire_{Biomathématiques}_Biophysique_Biochimie_et_de_{Scientométrie}_(L3BS),_Faculté_des_Sciences_de_la_Nature_et_de_la_Vie_Université_de_Bejaia,₀₆₀₀₀

Bejaia,Algeria

b_Faculté_des_Sciences_de_la_Nature_et_de_la_Vie_et_des_Sciences_de_la_Terre,_Université_de_Bouira,₁₀₀₀₀_Bouira,_Algeria

c_Laboratoire_{Biomathématiques}_Biophysique_Biochimie_et_de_{Scientométrie}_(L3BS),_Faculté_des_Sciences_de_la_Nature_et_de_la_Vie_et_des_Sciences_de_la_Terre,

UniversitédeKhemisMiliana,44225KhemisMiliana,Algeria

d_Technical_Sciences_Department,_Institute_of_Sciences,_University_of_Akli_Mohand_Oulhadj,_Bouira,_Algeria e_School_of_Food_and_Agriculture,_University_of_Maine,_Orono,_ME_04469,_United_States

a

r

t

i

c

l

e

i

n

f

o

Articlehistory: Received28May2015

Receivedinrevisedform24August2015 Accepted29August2015 Keywords: Pistacialeaves Ultrasoundextraction Phenoliccompounds Antioxidantactivity DOE

Artiﬁcialneuralnetworks

a

b

s

t

r

a

c

t

Designofexperiments(DOE)basedoncentralcompositedesign(CCD)andartificialneuralnetworks (ANNs)wereefficaciouslyappliedforthestudyoftheoperatingparametersofultrasoundassisted extrac-tion(UAE)intherecoveryofphenoliccompoundsfromP.lentiscusleaves.Thesemodelswereused toevaluatetheeffectsofprocessvariablesandtheirinteractiontowardtheattainmentoftheir opti-mumconditions.Undertheoptimalconditions(13.79minextractiontime,33.82%amplitudeand30.99 %ethanolproportion),DOEandANNmodelspredictedamaximumresponseof140.55and138.3452 mgGAE/gdw,respectively.Ameanvalueof142.76±19.98mgGAE/gdw,obtainedfromrealexperiments, demonstratedthevalidationoftheextractionmodels.Acomparisonbetweenthemodelresultsand experimentaldatagavehighcorrelationcoefficients(R2

ANN=0.999,R2RSM=0.981),adjustedcoefﬁcients

(RadjANN=0.999,RadjRSM=0.967)andlowrootmeansquareerrors(RMSEANN=0.37andRMSERSM=4.65)

andshowedthatthetwomodelswereabletopredictatotalphenoliccompounds(TPC)bygreen extrac-tionultrasoundprocess.TheresultsofANNwerefoundtobemoreconsistentthanDOEsincebetter statisticalparameterswereobtained.

1. Introduction

Nowadays,consumersarehighlyawareofthecloserelationship betweennutritionand healthand theywanttoinclude health-promotingingredientsintheirdiets.Therefore,naturalingredients recovered fromvascular plants have speciﬁc dietary and func-tionalpropertiesandcanbeutilizedeffectively todevelop food additivesorsupplementswithhighnutritionalvalue(Gruenwald, 2009). For example, the Food and Drug Administration (FDA)

∗ Correspondingauthorat:TargaOuzemour,Bejaia,06000.Algérie. E-mailaddress:farid.dahmoune@univ-bejaia.dz(F.Dahmoune).

classiﬁesferulicacidas anantioxidant inthe listoffood addi-tives (Lupien,2002).Wholefoodshave beenestimatedtohave between 5,000 and 25,000 individual phytochemicals (Thabit et al., 2015).Among these,phenolic compounds are secondary metabolites,ubiquitouswidelyexistinnatureandfood-industry by-products.They aredifferentiated fromone anotherby their structureandmolecularweight,andtheresulting physicochem-icalandbiologicalproperties.Duetothisenormousvariety,there arereportsofmore than10000phenolic moleculesandthelist continues expanding (Vázquez et al., 2015).Theyshow a large diversity of structures including simple phenols (C6); phenolic acidsandrelatedcompounds(C6–C1);acetophenonesandphenyl aceticacids(C6–C2);cinnamicacids,cinamylaldehydes,and

alco-http://dx.doi.org/10.1016/j.indcrop.2015.08.062

(3)

hols(C6–C3);coumarins,isocoumarins,andchromones(C6–C3); ﬂavonoids (C15); biﬂavonyls (C30); stilbenes (C6–C2–C6); ben-zophenonesandxanthones(C6–C2–C6);quinones(C6,C10,C14); betacyanins(C18);andlignans,lignins,tannins,andphlobaphenes (whicharedimmers,oligomers,orpolymers).

Allthesephenolicsarewelldocumentedandestablished,while theyexhibitgreatattentionduetotheirabilitytopromote ben-efits for human health such as the reduction in the incidence ofsomedegenerativediseaseslikecancer,diabetes,reductionin riskfactorsofcardiovasculardiseases,anti-inflammatory, antioxi-dant,antimutagenic,antimicrobial,anticarcinogenicactivitiesand helpinthepreventionofcytotoxiceffectsofoxidizedlow-density lipoprotein(LDL)andconsequentlyatherosclerosis(Delpino-Rius etal.,2015).Duetotheseomnipresenceandcountlessbeneficial characteristicsforhumanhealthaswellastherecentconsumer demandfor“allnatural”,researcheshavebeenintensifiedaimingto valorizationofplants,agriculturalby-productsandagro-industrial residuesasnatural-identicalingredientantioxidantadditivesand colorantsinsteadofsynthetic counterpartswhich proved nega-tivehealtheffects, suchas butylatedhydroxyanisole(BHA) and butylatedhydroxytoluene(BHT)(Dawidowiczetal.,2015).Pistacia lentiscusproductshaveawiderangeofusesinfoodindustrydueto activitiesrelatedtotheirsecondarymetabolites,suchasflavonoids, polyphenolsandphenolicacids(Bampoulietal.,2014).

Thisdiverserangeofbiologicalpropertiesmakesspice pheno-licsaninterestingtargetfortheirextractionconditions.Currently, extractionisbeingcarriedoutusingtraditionalprocessing tech-nologies including solvent extraction (both liquid–liquid and solid–liquidextraction)by assistanceofdifferentprocesses like mechanicalagitation,pressing,orheatingsystems.Theseprocesses arestillpopularandthemostregardedamongstallthe conven-tionalextractionmethodsbecausetheyhavebeenwellestablished andareeasytoperformandtooperate.However,thesemethods have beenassociated witheconomic impact, highsolvent con-sumption,longerextractiontimes(upto1500minormore)and anincreasedriskofthermaldegradationofsensitivecomponents (Spignoetal.,2007).

Theseshortcomingshaveledtotheconsiderationof theuse ofnew “eco-friendly” techniquesin separation, which typically uselesssolventandenergy(Chematetal.,2012).Inlastdecade, fewalternativetechnologiesasenergysavingincludingultrasound assisted extraction (UAE) (Dahmoune et al., 2014), supercriti-calﬂuid extraction(SFE) (Veggi etal.,2014), pressurizedliquid extractions(PLE)(Machadoetal.,2015)andmicrowaveassisted extraction(MAE)(Dahmouneetal.,2015)areused.

Fromtheliteraturereview,theuseofultrasoundisincreasingly employedasanalternativemethodoverthetraditionalin extrac-tionofnaturalproductsbecausetheyareshorteningofprocessing andresidencetimes(theycannowbecompletedinminutesinstead ofhourswithhighreproducibility)andacceleratedheatandmass transfer,eco-friendly,withstronglydecreasedsolvent consump-tionandqualityoftheextracts(Rami ´cetal.,2015).Inallcases,for keepingtheseadvantages,productionofnaturalfunctional ingredi-entsfromnaturalsources,represents:auniquelow-costresource. The increasingnumber of investigations publishedin this area highlightsunyieldinginterestsintheapplicationofmathematical modelsfortheantioxidants-makingoptimization-process.

Degree Of Experiment (DOE) based on Response Surface Methodology(RSM)isthecommonlyusedoptimisationtoolthat optimisestheprocessvariables duringtheextractionof natural phenoliccompoundsfromtheplantsources(Gongetal.,2012). RSMisoneoftheDOErelevantmultivariatetechniquesthatcan dealwithexperimentaldesign, statisticalmodelingand process optimization(Haﬁzietal.,2013b).RSMgenerallyappliesthe cen-tralcomposite design andBox–Behenken design (Dahmoune et al.,2013).Ithavebeenwidelyusedintheindustriestodevelop

their production as well as to optimise processes that have several factors that influence the responses of the final prod-uct.RSMisacollectionofmathematicalandstatisticalmethods to:thatevaluaterelationshipsbetweenagroupofindependent variablesand oneormoreresponses.CentralCompositeDesign (CCD) is one of RSM and is a model of second degree with several variables, it is easy to implement and has the prop-erty of sequentiality. We can undertake the study of the first k factors while reserving the option to add new without los-ingtheresultsof testsalready carried out.The useoftheCCD designispopularinresearchbecauseitisaneconomicaldesign. Apartfromtheconventionalstatisticalmethods(DOE),employing artificialintelligencebasedtechniquespredictionsuchasANNs, fuzzy inference systems (FIS), genetic programming (GP), and regression trees are recently highlightedin literature(Sarve et al.,2015).Recently, ANNs arean interesting methodin several Artificial Neural Networks:, it is one of the important artificial intelligencetoolswhichcanbeusedtosolveproblemsthatare noteligibleforconventionalstatisticalmethods(Shojaeimehret al., 2014). Some researchers tried to develop an evaluation of RSMand ANNs astwo usefulmethods topredict and simulate recovery of phenolic compounds using conventional processes from Garcinia mangostana (Cheok et al., 2012), and efficient extractionofartemisininfromArtemisiaannua(Pilkingtonetal., 2014).

Therefore,inthepresentinvestigation,a twolevelthree fac-tor(23₎_full_factorial_central_composite_design_(CCD)_in_RSM_and

anANNbasedmodelweredevelopedand comparedtoindicate howthepercentageofTPCrecovery,throughthegreenextraction processofP.lentiscusleaves,mightbeoptimized.The experimen-talvariablesincludingethanolproportion,amplitudeintensityand extractionduration,inanextractionprocedurewhichisconsidered toapproximatetheindustrialapplication.Thedevelopedmodels werethencomparedintheirsuitabilityforpredictingTPCrecovery byanalysingtheircoefﬁcientofdetermination(R2_),_adjusted

coef-ﬁcientofdetermination(R2_adj),_root_mean_square_error_(RMSE),

meansquareerror(MSE),absoluteaveragedeviation(AAD),Mean AbsoluteError(MAE),relativepercentdeviation(RDP)and stan-darderrorofprediction(SEP)fromexperimentaldata.Themodels werethenusedtodeterminetheeffectoftheextractionconditions onTPCrecovery,therebyprovidinganindicationoftheoptimal conditions.Finally,theoptimalsolutionsofferedbyRSMandANN werestatisticallycomparedtotheexperimentalvaluesbyusinga pairedt-test.Tothebestofourknowledge,thisistheﬁrstreport comparingRSMand ANNwithseveralstatisticalparameters,in naturalphenolicultrasoundextractiontechnologyfromP.lentiscus leaves.

2. Materialsandmethods

2.1. Plantmaterials

TheleavesofP.lentiscuswereharvestedinJune2012from spon-taneousplantsinOuedGhir,Bejaia,locatedintheNorthEastof Algeria.ThecollectedsampleswereidentiﬁedbytheVeg-etable EcologicalLaboratoryoftheAlgiersUniversity,Algeria.The sam-pleswerewashedwithtapwaterfollowedbydistilledwater,dried inastaticovenat40◦Cforaboutoneweek,andthengroundin anelectricalgrinder(IKAmodelA11Basic)toobtainﬁnepowder.

Then,thepowderwaspassedthroughstandard125␮msieveand storedinairtightbagsunderdarknessuntiluse.Moisturecontent wasassessedbyconstantweightat105◦Candwas5.0±0.5%,the wateractivityAw(HygroPalmAw)was0.18±0.2at20.6◦C.

(4)

Table1

TheexperimentaldesignlayoutandcorrespondingresponsesforCentralcompositedesign(CCD)basedonresponsesurfacemethodology(RSM)andartiﬁcialneuralnetwork (ANN)andpredictedvaluesforyieldoftotalphenoliccompoundsfromP.lentiscusleaves(TPC)usingultrasoundassistedextraction(UAE).GAE,gallicacidequivalents; referredtodryweight(dw).

Run X1-Time X2-Amplitude X3-Solvent Experimental TPCyield(mgGAE/gdw)

Predicted Residual RSM ANN RSM ANN 1 5 30 30 138.39 139.55 138.55 −1.16 −0.16 2 15 30 30 139.79 140.96 139.85 −1.17 −0.06 3 5 70 30 163.41 160.37 163.42 3.04 −0.01 4 15 70 30 110.57 108.78 110.48 1.78 −0.09 5 5 30 70 63.33 66.98 63.48 −3.65 −0.15 6 15 30 70 109.18 114.07 109.02 −4.89 0.16 7 5 70 70 157.51 158.20 157.66 −0.69 −0.15 8 15 70 70 151.60 152.29 152.30 −0.69 −0.70 9 10 50 50 142.46 143.49 142.25 −1.03 0.21 10 10 50 50 141.55 143.49 142.25 −1.94 −0.7 11 10 50 50 142.08 143.49 142.25 −1.41 −0.17 12 10 50 50 144.11 143.49 142.25 0.62 1.86 13 17.05 50 50 146.32 143.57 145.68 2.75 0.64 14 2.95 50 50 147.72 146.74 147.37 0.98 0.35 15 10 78.2 50 146.32 149.53 142.26 −3.21 4.06 16 10 21.8 50 114.85 107.90 114.96 6.95 −0.11 17 10 50 78.2 126.10 119.83 126.01 6.27 −0.09 18 10 50 21.8 137.77 140.31 138.01 −2.54 −0.24

2.2. ExtractionandquantiﬁcationofTPC 2.2.1. Ultrasoundassistedextraction

TheextractionofphenoliccompoundsoriginatingfromP. lentis-cusleavesbymeansofultrasoundwasperformedaccordingtothe CCDmatrixmodel.ACCDconsistsofthreesectionsincludingthe fullfactorialdesignpoints(wherethefactorlevelsarecodedtothe upperlevelthatcorrespondsto+1andthelowerlevelto−1 val-ues),axialpoints(sometimescalled“star”points,wherethefactor levelsarecodedtotheupperlevelthatcorrespondsto+␣andthe lowerlevelto-␣values)andthecentralpointthatcanbereplicated toprovideanestimationofexperimentalerrorvariance(Hafiziet al.,2013a)(seedetailsbelow,Section2.4).AccordingtotheDOE model,theUAEwasperformedatdifferentultrasoundpowerofthe transducer(21.8–78.2%), extractionduration(2.5–17.5min) and ethanolproportion(21.8–78.2%)atroomtemperaturewhichwere controlledbycirculationofacoldwater(25±2◦C)using thermo-staticpump(Table1).Ultrasonicirradiationwasappliedbymeans ofaUP-200Ssonifier(200W,24kHz)(HielscherUltrasonics, Tel-tow,Germany).TheextractwasfilteredthroughaWhatmanNo.1 (0.45␮m)filterpaperundervacuumpumpandthefiltratewas thendilutedtothefinalvolumeof50mLandwasusedforanalysis. After DOE–CCD-RSM and ANN optimization methods, the extractionofP.lentiscusleavespowderwasdoneunderoptimized operationalconditions.Onegramsofleavespowderwereextracted with40mLof30.99%ethanol,amplitudeintensityof33.82during 13.79min.Thenthefinalvolumeoftheextractwascompletedtoto 50mLwiththeextractionsolventandwasusedforfurtheranalysis.

2.2.2. Acceleratedsolventextraction(ASE)procedure

Extractionofthepistacialeavematerialswascarriedoutusing acceleratedsolventextractor(ASE,DionexCorp.,Sunnyvale,CA) system.Theleavepowder(1g)wasplacedintwolayersof diatoma-ceousearth(about0.5gineachlayer)in11mLDionex(ASE200) stainless-steelcell,andextractedwithethanol40%.Thecellswere equippedwithastainlesssteelfritandacelluloseﬁlter(Dionex Corp.)atthebottomtoavoidthecollectionofsuspended parti-clesinthecollectionvial.Adispersingagent(diatomaceousearth), wasusedtoreducethesolventvolumeusedfortheextraction. Theextractionwasperformedat1500psiandoptimal tempera-ture(120◦C),andthenheatedfor6min,followedbyathreestatic

periodsof5min(3staticcycles).Thesamplewasﬂushedwith70 %andpurgedwithaﬂowofnitrogenduring90s.Extracts(25mL) werecollectedinto50mLtubes.Theextractwasthenrecovered andanalysedasreportedinSection2.3.1fortheoptimizedUAE extract.

2.2.3. Conventionalextraction

Fortheconventionalsolventextraction,1gofﬁnepowderwas placedinaconicalﬂaskof250mL(Ø×H:51×150mmandcapsize of38mm),and50mLof40%(v/v)EtOHwereadded.Themixture waskeptinathermostaticwaterbath(mod.WNB22,Memmert) withshaking speed of110strokes perminute, at 60◦C for 2h, accordingtothemethodrecommendedbySpignoetal. (2007). TheextractwasthenrecoveredandanalysedasreportedinSection 2.3.1fortheoptimizedUAEextract.

2.2.4. Determinationofpolyphenolcontent

ThecontentinTPCoftheP.lentiscusleavesextractwas deter-minedbyFolin’sassayasreportedinliterature(Jaramillo-Flores etal.,2003).Brieﬂy,100␮Loftheextractwasmixedwith750␮L Folin–Ciocalteudilutedreagent(1/10).Then, thesolutions were mixedandleftatroomatroomtemperaturefor5min.Afterthat, 750␮Lof7.5%sodiumcarbonate(Na2CO3)solutionwasadded.

Afterincubationat25◦Cfor90mintheabsorbancewasmeasured at725nm(1cmopticalpath)againstablank(madeasreportedfor thesamplebutwith100_␮Lofsamplesolvent)usingaSpectroScan 50UV–visspectrophotometer.Gallicacidhydratewasusedas stan-dardforthecalibrationcurvetoexpresstheTPCconcentrationof thesampleasmgL−1ofgallicacidequivalents(GAE).TPCyieldwas thencalculatedbasedonthesampleconcentration,extractvolume andleavespowderdryweight,accordingtothefollowingequation (Eq.(1)): TPyield= mgGAE L×LExtract gdwleavespowder (1) 2.3. Antioxidantpower–scavengingeffect

2.3.1. ScavengingactivityagainstDPPH•radical

Thefreeradicalscavengingabilityoftheextractwasmeasured usingacolorimetricmethodwherechangeinthepurplecolour solutionofDPPH(1,1-diphenyl-2-picrylhydrazyl)radicalwas

(5)

mea-suredonaplatereader(OmegaFLUOstar,BMGLabtech,Cary,NC, USA).300␮LofDPPH•solutioninmethanol(70␮M)wasmixed with10␮Lofextractandthemixturewasincubatedfor20min at37◦C.Thedecreaseinabsorbancereadingofthemixturewas measuredat515nm.Theantioxidantcapacityoftheextractwas expressedasapercentageofinhibitionofDPPHradical(% inhibi-tionofDPPHradical)calculatedaccordingtothefollowingequation (Eq.(2)):

%inhibition=AOX=Ablankt−20Asamplet−20 ADPPHt−0

(2) where,Ablankistheabsorbancevalueoftheblank(300␮LofDPPH

solutionplus10␮Lofthesolventinwhichtheextracthasbeen dissolved);Asample,istheabsorbanceofthesampleextract;tisthe

time(min)atwhichabsorbancewasreadandADPPHisabsorbance

ofthecontrolattime=0min.Theeffectiveconcentrationof sam-plerequired to scavenge DPPH radical by 50 %(IC50 value) is

obtainedbylinearregressionanalysisofdose–responsecurve plot-tingbetween%inhibitionandconcentrations.

2.3.2. Oxygenradicalabsorbancecapacity(ORAC)assay

Theantioxidantcapacityoftheextractwasalsoperformedusing anoxygenradicalabsorbancecapacity(ORAC)assayusinga flu-orescenceplatereaderfollowingtheproceduresofHuang etal. (2002)withslightmodifications.Theplatereaderwasequipped withanincubatorandtwoinjectionpumps.Briefly,sampleextracts of20␮Lwithsuitabledilutioninphosphatebuffersalinesolution (75mM,pH7.0)wereloadedtopolystyrene96-wellmicroplatesin triplicatebasedonarandomisedsetlayout.

Thesamplesweredilutedtotheproperconcentrationrangefor fittingthelinearityrangeofthestandardcurve.Afterloading20␮L ofsample,standardandblank,and200␮Lofthefluorescein solu-tionintoappointedwellsaccordingtothelayout,themicroplate (sealedwithfilm)wasincubatedfor15minintheplatereaderat 37◦Candthen20␮Lofperoxylgenerator2,2 -azobis(2-amidino-propanedihydrochloride(AAPH)(3.2␮M)wasaddedtoinitiatethe oxidationreaction.Theplatereaderwasprogrammedtoinjectand recordthefluorescenceoffluoresceinoneverycycle.Thekinetic readingdecreaseofFluoresceinintensity(%)wasrecordedfor60 cycleswith40spercyclesettingduring40minat37◦C.Absorbance readingsoftheplateweretakeneverycycleusinganexcitation wavelengthof485nmandanemissionwavelengthof535nmuntil allfluorescencereadingsdeclinedtoless than25%oftheinitial values.

FourTroloxsolutions(6.25,12.5,25,50␮Minphosphatebuffer salinesolution(75mM,pH7.0)wereusedtoestablishastandard curve.

Theareaunderthecurve(AUC)wascalculatedforeachsample byintegratingtherelativeﬂuorescencecurve(Eq.(3)).

AUC=

0.5+f4 f3+ f5 f3+ f6 f3+ ...+ fi f3

×CT (3)

wheref3istheinitialﬂuorescencereadingatcycle3,fiisa

ﬂuores-cencereadingatcyclei,andCTiscycletimeinminutes.Thenet AUCofthesamplewascalculatedbysubtractingtheAUCofthe blank.

TheregressionequationbetweennetAUCandTrolox concen-trationswasdeterminedandORACvalueswereexpressedas␮mol Troloxequivalentspergramofsample(␮molTEg−1)usingthe standardcurveestablishedinthesameconditions(R2₌_0.9946).

2.4. Mathematicalmodels

2.4.1. Centralcompositedesign(CCD)

The effects of extraction processing variables on phenolic contentsincludingethanolproportion,extractiontimeand

extrac-tionamplitudewere investigatedby theone-variable-at-a-time method(preliminarystudy).Onthebasisofthesinglefactor exper-imentalresults,majorinﬂuentvariableswereconﬁrmed,andthen aresponsesurfacemethodologywasconductedtodesign exper-imental project. JMP software package was used toestablish a CCD-RSMmathematicalmodelandobtaintheoptimumconditions ofthegreenultrasoundtechnology.

TheCCDconsistof:

• FullFactorialDesign(FFD)withtwolevels(nf)(codedvalues±1):

nf =2k,wherekisthenumberoffactors(threevariablesinthis

work)(Table1);

• Centralpoints,couldbeafunctionoftheexperimenterandhave thecoordinatesn0(codedvalues0).Theseexperimentsareused

toassesstherepeatability,theyareusedtoensurethatthereis noslipbetweentheFFDandstarplan,theyareinvolvedinthe calculationof␣,thesepointsprovidealsoameanforestimation oftheexperimentalerrorandprovideameasureoflack-of-ﬁtand ﬁnallytheyareusedtotestthevalidityofthemodel(Table1); • The“starpoints”(n␣=2k)(codedvalues±␣),addtwomore

lev-elsineachkvariable(Table1).Thetotalnumber(n)oftheCCD tobeperformedis(Eq.(4)):

ntotaloftest=nf+n˛+n0 (4)

The␣valueiscalculatedtoobtainanear-orthogonality. Fre-quentlythevalueof␣isselectedtomaketherotatabledesignand issatisﬁedasfollow(Eq.(5)):

˛=

⎛

⎜

⎝

nf

n0+nf+n˛−

nf

2 4

⎞

⎟

⎠

1 4 (5)

Thevariableswerecodedaccordingtothefollowingequation (Eq.(6)):

xi=Xi−X0

X (6)

whereXiisthe(dimensionless)codedvalueofthevariableXi,X0is

thevalueofXatthecenterpointandXisthestepchange. TheexperimentdesignisshowninTable1,alongwith exper-imental data. The extraction rate of TPPL is the response. The regressionanalysiswasperformedtoestimatetheresponse func-tion as a second-order polynomial, the variables were coded accordingtothefollowingequation(Eq.(7)):

YTPPL=B0+ k

i=1 BiXi+ k

i=1 BiiX2+ k

ij BijXiXj+E (7)

where,YTPPL representedtheresponsefunctionoftotalphenolic

fromPistacialeaves(TPPL).B0wasconstantcoefﬁcient.Bi,BiiandBij

werethecoefficientsofthelinear,quadraticandinteractiveterms, respectively.Accordingtothestandardleastsquaremethodsand analysisofvariance,theregressioncoefficientsofindividual lin-ear,quadraticandinteractiontermsweredetermined.Inorderto visualizetherelationshipbetweentheresponseand experimen-tallevelsofeachprocessingvariableandtodeducetheoptimum conditions,theregressioncoefficientswerethenusedtomake sta-tisticalcalculationtogenerate3-D surfaceplotsfromthefitted polynomialequation.TheP-valuesoflessthan0.05,0.01and0.001 wereconsideredtobestatisticallysignificant.

Thedeterminationcoefﬁcient(R2_),_adjusted_coefﬁcient_(R2_adj),

therootmean-squar-error (RMSE),adequationprecesion (Adeq prec)ofthefunctionwereusedforevaluatingthestatistical sig-niﬁcanceofthemodeldeveloped(Table2).

(6)

Table2

Analysisofvariance(ANOVA)fortheexperimentalresultsobtainedbyusing ultra-soundassistedextraction.

Source Sumofsquares DF*** _F-value _P-value_Prob>F

Model 8832.36 9 47.33 <0.0001 X1-Time 9.02 1 0.43 0.5306 X2-Amplitude 2585.17 1 124.67 <0.0001 X3-Ethanol 620.26 1 29.91 0.0009 X1X2 1424.97 1 68.72 <0.0001 X1X3 1061.45 1 51.19 0.0002 X2X3 2451.05 1 118.2 <0.0001 X12 19.63 1 0.94 0.363 X22 371.52 1 17.91 0.0039 X32 303.52 1 14.63 0.0065 Residual 145.14 8 LackofFit 142.67 5 23.06 0.0421 PureError 2.47 3 R2 _0.98 R2_adj _0.96 Adeqpre* _26.25 RMSE** _4.65 CorTotal 9165.69 18 *_Adequation_precision. **_Root_mean_square_error. ***_Degree_of_freedom.

2.4.2. Artiﬁcialneuralnetworkartiﬁcial(ANN)

Artificial neural networks (ANNs) are mathematical models, whichwereinspiredfromthefunctioningof biologicalnervous systemsandcomefromArtificialIntelligence,haveprovedtobe apowerfulmodelingtechniqueforcomplexandnonlinear prob-lemswithstrongproposedtolearningandfunctionapproximation (Hafizi etal., 2013b).Theexperimentaldataontherecoveryof phenoliccompoundsobtainedbyUAEmethodwasusedfor train-ingtheANNanditwasdevelopedinJMP.Amultilayerperceptron (MLP)isafeed-forwardANNconsistingthreeormorelayersof neu-rons,withthefirstlayerofneuronsrepresentingtheindependent parameterinputs(Fig.1).Eachoftheneuronsinthefirstlayeris connectedtooneormorelayersofhiddenneuronsthatrepresent nonlinearactivationfunctions(Pilkingtonetal.,2014).These neu-ronsareinturnconnectedtoafinallevelofoutputneuronsand, throughtheseoflearningalgorithms,therelativeinfluenceofeach inputneuronandtheircomplexinteractionsontheobservedresult canbediscerned(Pilkingtonetal.,2014).TheinputstotheANN modelwereidenticaltothefactorsconsideredinRSMapproach, namelyextractiontime,amplitudeandethanolproportion,a sin-glehiddenlayerofneurons,andanoutputneuronrepresentingthe recoveryoftotalphenoliccompounds,wasasimilartoRSM model-ing(YTPC.ArepresentationoftheMLParchitecturecanbeobserved

Fig.1.ANNdesignforthreeoperatingparameters(irradiationtime(X1);Amplitude

(X2)andethanolproportion(X3))asinput,andTPCyieldasoutput.

inFig.1.Thenumberofhiddenneuronsrequiredinthehidden layerwasdeterminedbytrialanderrortominimisethedeviation ofpredictionsfromexperimentalresultsplannedthroughCCD.A totalof18ofexperimentalresultswereusedtotrainthenetwork, withtheremainingresultssplitevenlybetweennetworkvalidation andtesting(Table1).Inourcase,K-Foldcross-validationprocedure wasusedforthevalidationandtestingofANN-model,consisting ondividingtheoriginaldataintoKsubsets.Inturn,eachoftheK setsisusedtovalidatethemodelfitontherestofthedata,fitting atotalofKmodels.Themodelgivingthebestvalidationstatistic ischosenasthefinalmodel(Table3).Thismethodisthebestfor smalldatasets,becauseitmakesefficientuseoflimitedamounts ofdata.TheANNpredictionswerethenusedtogeneratesurface plots.

Thedeterminationcoefﬁcient(R2_),_the_root_{mean-squar-error}

(RMSE),Loglikelihoodtest,MeanAbsoluteDeviation(MAD)and sumofsquarerror(SSE)ofthefunctionwereusedforevaluatingthe statisticalsigniﬁcanceofthetheANNsmodelsdeveloped(Table3).

2.5. Statisticalanalysis

ThestatisticalsoftwarepackageJMP(10.0.0version,SAS Insti-tute) was used for a regression analysis of experimental data, plottingtheresponsesurfacesandanalysisofvariance(ANOVA) ofthestatisticalparametersforTPCrecoveries.

(7)

Table3

PerformanceofANNtrainingandANNvalidationmodels.

TPC(mgGAE/gdw) Training

R2 _0.999741

RMSE* _0.3722593

MAD** _0.2902812

-LogLikelihoodtest 6.4616106

SSE*** _2.0786553 SumFreq 15 Validation R2 _0.9948071 RMSE 1.0740733 MAD** _0.6852675 -LogLikelihood 4.4711903 SSE*** _3.4609004 Table4

Comparisonoftheperformanceofresponsesurfacemethodologyandartiﬁcial neu-ralnetworks.

Performanceequations RSM ANN

R2₌

N i=1(Xi−Yi) 2

N i=1

¯Yi−Yi

2 0.981 0.999 R2_Adj₌₁₋

₁₋_R2

_× N−1 N−k−1

0.967 0.999 SEP=RMSE Ye ×100 3.45 0.28 MAE= N

i=1

_Y i,exp−Yi,pred N

0.0006 0.26 AAD=

1 N N

i=1

_Y i,pred−Yi,exp Yi,exp

×100 0.14 0.18 RDP=100 N N

i=1 |

Yi,pred−Yi,exp

| Yi,exp 0.103 0.022 MSE=

N

i=1(Yi,exp−Yi,pred)2

N 9.63 1.21

RMSE=

N

i=1(Yi,exp−Yi,pred) 2

N 4.65 0.37

MSE,meansquarederror;RMSE,rootmeansquareerror;AAD,absoluteaverage deviation;MAE,meanabsoluteerror;RDP,relativepercentdeviation;SEP,standard errorofprediction.

Inﬂuenceofextractiontechnique(UAE,ASEorCSE)onTPCyield andextractAOCwasassessedbyunivariateANOVAandTukey’s posthoctestformeansdiscrimination(95%conﬁdencelevel).

In addition, to evaluate the goodness of fit and prediction accuracyoftheconstructedmodels,correlationcoefficients(R2_), adjusteddeterminationcoefficient(R2_adj),_root_mean_square_error (RMSE),MeanSquaredError(MSE),MeanAbsoluteError(MAE)and absoluteaveragedeviation(AAD),relativepercentdeviation(RDP), standarderrorofprediction(SEP)werecarriedoutbetween exper-imentalandpredicteddata(Table4)(Rafighetal.,2014;Sarveet al.,2015).

3. Resultsanddiscussion

3.1. ModelingandoptimizationusingCCD–RSMofextraction process

AccordingtotheCCD,experimentswereperformedinorderto ﬁndouttheoptimumcombinationconditionsandstudytheeffect ofprocessparametersonTPCrecoveryusingtheultrasound pro-cess, andtheresultsare givenin Table1. In thepresentwork, therewereatotalof18runsforoptimizingthethreeprocessing

parameters.Byapplyingmultipleregressionanalysisonthe exper-imentaldata,themathematicalmodelrepresentingtherecovery ofTPCUAEwithinthelevelsunderinvestigationwasexpressedby

thesecond-orderpolynomialEq.(8)usingthestandardleastsquar (LSM)methodwithactualfactors:

Y1 =149.02+1.63X2−2.11X3−0.13X1X2+0.13X1X3

+0.043X2X3−0.02X22−0.01X32 (8)

Table2showstheresultofanalysisofvarianceforthefitting model.Valuesofprobability(P)>Flessthan0.05and0.01indicate thatmodeltermsaresignificantandhighlysignificantrespectively, andthevalues greaterthan0.05 indicatethat themodel terms arenotsignificant.FromtheTable2,theANOVArevealsthatthe modelwashighlysignificantwithap-value<0.0001,whichmeans thatthemodelrepresentedthedatasatisfactorily.Inotherways, 98.12%ofthevariationscouldbecoveredbythefittedmodel. Fur-thermore,thevaluesofpureerror2.47islowwhichindicategood reproducibilityofthedata.Thecoefficientsofdetermination(R2₎

andadjustedcoefﬁcients(R2_Adj)_are_close_to_1,_which_revealed

also,thatthereareexcellentcorrelationsbetweenthepredicted andexperimentalmodels(Fig.2a).Inaddition,thesmallvaluesof CV(3.45%)than10andhighervaluesofadequateprecisionthan4 forbothextractionplantsgivebetterreproducibilityandadequate signal(Table2).Moreover,“The ¨‘‘PredR-Squared” öf0.8422isin reasonableagreementwiththe ¨‘‘AdjR-Squared” öf0.9630;ï.e.the differenceislessthan0.2.Inconclusion,thesemodelscanbeused tonavigatethedesignspace.

3.2. Responsesurfaceanalysis

Theinteractionsamongdifferentprocessparametersand opti-mumidentifyingvaluesforreachingthemaximumrecoverywere investigatedbya3DsurfaceplotaccordingtoEq.(8).

Theplotsweredesignedbysketchingtheresponse(z-axis) ver-sustwooutputindependentvariables(xandycoordinates)and theotherindependentvariableswereconsideredconstantattheir zerolevelsofthetestingranges(GhoreishiandHeidari,2013).The recoveryofTPC mainlydepends ontheextractiontime, ampli-tudeandethanolproportionasitslineareffectforamplitudeand ethanolproportionandallmutualeffectswerehighlysignificantat p<0.01(Table2).TheP-valuesat0.05and0.01wereusedasatool tocheckthesignificanceorhighlysignificanceofeachcoefficient. ItcanbeseenfromTable2that,thelinearcoefficients(X2,X3)a

quadratictermcoefﬁcient(X22,X32)andcrossproductcoefﬁcients

(X1X2,X1X3,X2X3)(Fig.3),werehighlysigniﬁcant,withverysmall

P-values(P<0.01).Theothertermcoefﬁcientswerenotsigniﬁcant (P>0.05).

InFig.4A,whenthe3-Dresponsesurfaceplotwasdevelopedfor theextractionyieldofTPCwithvaryingextractiontimeand out-putamplitudeatfixedextractionsolvent(0level),theextraction yieldofphenoliccontentincreasedwiththeincreasingextraction time(2.5–17.5min),andandwasenhancedrapidlywithincrease ofoutputamplitudefromabout30–78%overinvestigatedrange (Fig.3).ItcanbeseenfromTable2that,thecrosseffectof time-amplitude(X1×2)washighlysignificant,withverysmallp-values (P<0.0001). The intensification of extraction process has been attributedtothephysicaleffectsofultrasoundsuchasacavitation whichcanleadtoasignificantincreaseinthemasstransferrates. Ultrasoundcanbeeffectivelyusedtoincreasetheyieldandrateof masstransferinseveralsolid–liquidextractionprocesses.Similar resultsobtainedbyKadametal.(2015)alsoshowedthat ultrason-icationcouldincreasetherecoveryofbioactivecompoundsfrom Ascophyllumnodosummaterials.

(8)

Fig.3. Interactionplotsoftheprocessingvariables(toirradiationtime(X1);Amplitude(X2)andethanolproportion(X3))ontherecoveryofTPC.Theinteractionshowsas

nonparallellines.Redlinesdisplayfactoratlowlevel,whereasbluelinesarethehighlevelofthefactors.(Forinterpretationofthereferencestocolourinthisﬁgurelegend, thereaderisreferredtothewebversionofthisarticle.)

Fig.4BdepictstheextractionyieldofTPCasafunctionof extrac-tiontimeandaqueousethanolatﬁxedoutputamplitude(0level or50%),itindicatedthatthemaximumextractionrecoveryofTPC canbeachievedwhenextractiontimeandethanolproportionare atthelevelsof2.5minand20%,respectively(Fig.3).Itisclearthat theinteractionbetweentheseinvestigatedvariabesishighly sig-niﬁcantwithp<0.0002.Infact,inultrasoundprocess,acavitation phenomenaproducedathighamountofwater(80%,lowpurityof extractionsolvent)isoneofthekeyvariablesaffectingthereleaseof polyphenolsfromdifferentmatricesbyviolentcavitation collaps-ingbubbles,whichhavetheabilitytomodifyequilibriumandmass transferconditionsduringextraction.Duetocavitation,thecracks aredevelopedinthecellwallwhichincreasespermeabilityofplant tissuesfacilitatingtheentryofthesolventintotheinnerpartofthe materialaswellaswashingoutoftheextracts(Rodriguesetal., 2015).Exposureofpolyphenolstotheultrasonicwavesforalonger periodbeyondabout5minresultsinthestructuraldestruction, whichresultsinareducedyield.Inaddition,increasingradiation timeinallreactions,includingtheformationoffreeradicalswhich canbescavengedbyphenoliccompounds(Jermanetal.,2010),are promotedwhenhightimes/amplitudesareused.

Theeffectofmutualinteractionbetweenoutputamplitudeand ethanolproportionisshowninFig.4Candwashiglysigniﬁcant (p<0.0001)withamaximumTPC yieldat78%ethanol propor-tionand78%amplitude(Fig.3).Resultsindicatedthatextraction ofactivecompoundsfromtheplantmaterialswereaccomplished after10min.Increaseintheethanolproportionrequiredhigh son-icationintensitytogeneratethecavitationbubbles.Itisknowthat thecavitationphenomenaisfavoratedbythedegreeofimpurities inthemedium.Thecavitationbubblesismainlymediumnature dependent.

Inconclusion,theoptimumofTPCrecoveryof140.55mgGAE g−1 wasobtainedviathecanonical analysisofresponsesurface

whichpredictedthefollowingconditionsas13.79minextraction time,33.82%amplitudeand30.99%ethanolproportion.The accu-racy of themodeling optimalTPC recovery wasvalidated with triplicateexperimentsgivingtheaverageextractionrecoveryof 142.76±19.98mgGAEg−1.

3.3. Artiﬁcialneuralnetworkmodel

TherecoveryofTPCaspredictedbytheANNiscomparedtothe experimentallyobtainedvaluesinTable1.Inordertotestthe suit-abilityofthemodel,thepredictedandactualresultswereplotted inFig.2b.

BysupplyingtheANNmodelwithmatricesofextraction con-ditionparameters,itwaspossibletovisualisetherelativeimpact ofeachextractionvariablesusingsurfaceplotgeneratedinJMP (Fig.5).UsingthesameapproachaswiththeRSMmodel,thethird variableforeachplotwasfixedatitsmiddlevalue(10min,50%or 50%)togeneratethesurfaceplotsandtheresultsofthis investi-gationcanbeobservedinFig.5.ItisdirectlyobviousfromFig.5A that,increasingtheultrasoundintensityfrom20%(code=−1)to 70%(code=+1)improvetherecoveryofphenoliccontentandthis leadstohigherinternalmasstransferandconsequentlyenhanced theTPCextractionrecovery.Extractionamplitudehasasignificant effectontherecoveryefficiencyofTPCasastateofequilibrium hasbeenattainedatthehighamplitude.Inthiscircumstance(with ethanol50%),theonlywaytoincreasetheextractionofpolyphenols istoincreasetheultrasonicationintensityofextraction,whichisin linewiththeincreasingsolubilityofphenoliccompoundsin extrac-tionsolventswithincreasingirradiationintensity.Theresultsin Fig.5B,indicatethattheinteractioneffectofextractionduration andethanolproportionisinversementproportionaltotherecovery ofTPC,whichsuggeststhattheextractionhasruntocompletion at30%ofethanolforabout5minandthathigherTPCrecovery

(9)

Fig.4.ResponsesurfaceanalysisforthetotalphenolicyieldfromP.lentiscusleaves withultrasoundassistedextractionmethodwithrespecttoirradiationtime(X1);

Amplitude(X2)andethanolproportion(X3),accordingtotheRSMmodel.

couldonlybeobtainedbyreducingtheconcentrationofethanol proportionandtime.

TheresultsinFig.3Cindicateanusualtrendathigherethanol proportioninwater,particularlybeyond60%,thehigherenergyof ultrasonicationrequired.Thisexplainedbythefactthatthe extrac-tionofTPCfromvegetabletissuematerialrequiresmoreenergy

Fig.5.ResponsesurfaceanalysisforthetotalphenolicyieldfromP.lentiscusleaves withultrasoundassistedextractionmethodwithrespecttoirradiationtime(X1);

Amplitude(X2)andethanolproportion(X3),accordingtotheANNmodel.

intensity,whenhighpercentageofpuresolventisusedtogenerate togenerateacavitationphenomena.Thisisanimportant observa-tionasitsuggeststhatproportionofsolventinwaterextractions withhighpurity maybeabletoretard acavitationin termsof extraction efficiency. As well as cost savings by running the extraction at lower proportion of organic solvents in water, there is the potential that the extraction of certain impurities (as chlorophyll, fibers, ...) may be reduced, lead-ing to a cleaner extract to be taken forward for purification.

(10)

-6 -4 -2 0 2 4 6 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 RSM ANN

Res

idual

Experiment

Numbers

Fig.6.ComparisonbetweentheresidualerrorsobtainedbyRSM(a)andANN(b)models.

It is hypothesized that a critical concentration of ethanol proportion is attained at about 40 % and that the solubil-ity of TPC is sufﬁciently improuved in the extraction mixture as to promote increased recovery from the Pistacia leaves. 3.4. ValidationandcomparisonofRSMandANNmodels

In the last decades, response surface methodology (RSM) andartificialneuralnetwork(ANNs) hasbecomethemost pre-ferred methods for machining (Kumar and Chauhan, 2015). Inthispresentwork,RSMandANNmethodswereappliedfor modeling and optimizationof ultrasound-assisted extractionof phenolicnaturalcompoundsfromP.lentiscusleaves. Inorderto validatetheadequacyofthemathematicalmodels,averification experimentcombinationfactorswascarriedoutunderthe opti-malconditions. Under theseoptimal conditions,RSM and ANN modelspredictedamaximumresponseof140.55and138.34mg GAEg−1dw,respectively.Ameanvalueof142.76±19.98mgGAE g−1dw,obtainedfromrealexperiments,demonstratedthe vali-dationof the extractionmodels.The good correlation between these results confirmed that the models were adequate for reflecting the expected optimization. The performance of the constructed ANNand RSM models werealso statistically mea-suredandsummarizedinTable4.Thedeterminationcoefficient (R2₎ _in _RSM _and _ANN _were _given _98.6 _% _and _99.9 _%,

respec-tively.Thus thecalculatedR2 _demonstrates_a _superior_accuracy

of ANN in contrast with the traditional RSM (Fig. 5). From a Table4,thestatisticalanalysisofRSMandANNmodelsprovided goodqualitypredictions,yettheANNshowedaclear superior-ity over RSM for both data ﬁtting and estimation capabilities. The actual and predicted values together with the residu-als, for both approaches are presented in Table 1. In order to compare the distribution of predicted and actual val-ues of two approaches, the results were plotted in Fig. 6. Theﬂuctuationsoftheresidualsarerelativelysmallandregular forANNcomparedtoRSM.TheRSMmodelshowsgreater

devia-Table5

ComparisonofextractionyieldandantioxydantactivitiesofpolyphenolsfromP. lentiscusleavesbyUltrasoundassistedextraction(UAE),Acceleratedsolvent extrac-tion(ASE)andConventionalsolventextraction(CSE).Resultsareexpressedas means±standarddeviation.

Antioxydantactivity

TPC(mgGAE/gDW) ORAC(␮molTE/gDW) DPPH(IC50)(␮gGAE/mL) UAE 142.76±19.98ab _517.52_±_35.18b _18.74_±_0.284c ASE 120.78±16.00b _257.07_±_20.00c _32.77_±_2.12a CSE 178.00±19.80a _671.07_±_58.80a _19.76_±_0.06b

Samelettersinthesamecolumnrefertomeansnotstatisticallydifferent(p>0.05).

tionthantheANNmodel.ItistobenotedthatthoughRSMhas theadvantageofgivingaregressionequationforpredictionand showingtheeffectofexperimentalfactorsandtheirinteractions onresponseincomparisonwithANN.ANNdoesnotrequirea stan-dardexperimentaldesigntobuildthemodel(Geyikc¸ietal.,2012). TheANNapproachisﬂexibleandpermitstoaddnew experimen-taldatatobuildatrustablemodel.Thusitwouldbemorerational andreliabletointerpretultrasound-assistednaturalphenolic com-pounds extractiondata through a process of ANNarchitecture. 3.5. ComparisonofUAEwithASEandtraditionalextraction

method

Oneof thequick methodstoevaluateantioxidantactivityis thescavengingactivityonDPPH,astablefreeradicalandwidely usedindex. TheDPPHmethodwasevidentlyintroducednearly 50 years ago by Blois(Molyneux,2004), and it is widelyused to test the ability of compounds to act as free radical scav-engersorhydrogendonors,andtoevaluateantioxidantcapacity. Results showed that antiradical activity was methods-dependent (Fig. 7 and Table 5). From the point of view of the quality of the extracts, the DPPH test showed UAE as the best technology due to the signiﬁcantly lower IC50 (19.76±0.06) in comparison to ASE and CSE (p<0.05).

(11)

0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30 35 40

% I

nh

ib

iti

on

D

PP

H r

adi

ca

l

mL Extract/

L

CSA UAE ASE

Fig.7. Reducingcapabilitiesofultrasound-assistedextracted(UAE)samplescomparedtoaccelerated-solventextraction(ASE)andconventional-solventextraction(CSE). Reducingcapacityoftheextractedwasquantiﬁedas%inhibitionusingDPPHradicals.

0 50000 100000 150000 200000 250000 300000 0 5 10 15 20 25 30 35 40

% F

luo

re

sc

en

ce

I

nt

ens

it

y

Time (min)

UAE CSE ASE

Fig.8.Fluorescenceintensitydecayofultrasound-assistedextracted(UAE)samplescomparedtoaccelerated-solventextraction(ASE)andconventional-solventextraction (CSE).ThiscurvewasusedtodeterminetheantioxidantactivityofP.lentiscusextractsusingORACassay.

From the point of view of industrial application of ultra-sound technology for separation as well as for the recovery of high-added value compounds from different plant materi-als,it is important toknow which factors influenceextraction yield, but above all it is important to develop mathemati-cal models that are able to describe and control the process. IntheORACassay,thedifferentextractsshowedthatthe antiox-idantpowers werestatistically different for UAE, CSE and ASE usingANOVA and Turkey’s post-hoc test with95 % confidence level(Table5).Theextracts-ASEshowedthatthelowest antiox-idantpowers;thesametendencywasobtainedwithDPPHtest. Fig.8depictsthetimecourseofthereactionoffluoresceinwith AAPHinthepresenceofdifferentextractsundertheconditions reportedinSection2.Itis alsoobservedthatASE-extractshave thelowest protection of fluorescence intensity during the test compared tootherextraction methods,beyond 15minthe flu-orescencesignalwasfoundtodecline dramaticallyin presence ofperoxyl generator(AAPH).However, byusing UAE and CSE-extractsthetendencywasprolongeduntilabout30min(Fig.8). The lowest antioxidant activities of ASE-extracts can be explained byunfavorable conditions of extractionas likeother

techniques. ASE is considered as a potential alternative tech-nique to conventional atmospheric pressure methods, for the extractionof naturel compounds. Compared with conventional solventextraction, there is a dramatic decrease in the amount of solvent and extraction time required for ASE; from our point of view of industrial and/or laboratory application of ASE mechanism for extract preparation as well as for the recovery of high-added value compounds from plant materi-als,it is important to know which factors inﬂuence extraction yield and it is important to develop mathematical models that are ableto describeand control the optimizationprocess. It was mentioned earlier that the TPC of extracts obtained usingUAEwasalsosimilartothat foundwiththeCSEmethod althoughthereweredifferencesintherequiredextractiontime usedinthetwootherprocedures.Inaddition,theoptimal extrac-tiontimerequiredformaximumORACactivityusingUAEwasjust 13minversus120minwithCSE,wesuggest here,thata longer extractiontime(approximatively10times)couldinﬂuenced neg-atively therecovery of bioactive compounds. The UAE process iscapableofextractingmorebioactivecompoundsinlesstime.

(12)

4. Conclusion

The capability of TPC extracted from pistacia leaves using aqueuse ethanol solution and the effects of process-ing parameters were investigated using RSM and ANN. The most important reveals from this investigationare as follows: • The optimum conditions were found as extraction time of 13.79min,33.82%amplitudeand30.99%ethanolproportion, andtheoptimumTPCrecoveryof140.55mgGAEg−1;

• Theresultsof ANNandRSMmodelsbasedonvalidationdata showedthatRSM(R2₌_0.986)_and_ANN_(R2₌_0.999)_are_useful_and

perfectmethodstopredictTPCbyapplyingUAEprocess.Itshows thatANNmodelisbetterthanRSM;

• StrongantioxidantactivityforUAEextractwereobtainedwith DPPHandORACradicalscomparedtotheASEtechnique,UAE allowsthequantitativeandreproducibleextractionofthe pheno-liccompoundspresentinPistacialeaves,inashorttimefollowed byCSE;

• TheresearchﬁndingsforUAEoptimizationwithRSMandANN modelswillprovideeffectiveguidelinesandtheresultswould beagooddatabasetotheFood-industryapplicationsforusein health-carefood;

• TheleavesofP.lentiscusarerichinpolyphenols.However, fur-therstudiesconcerningthenutritionalandhealthbeneﬁtsare requiredbeforealargescaleutilizationofthelentiscleavescan berecommended.

References

Bampouli,A.,Kyriakopoulou,K.,Papaefstathiou,G.,Louli,V.,Aligiannis,N., Magoulas,K.,Krokida,M.,2014.Evaluationoftotalantioxidantpotentialof Pistacialentiscusvar.chialeavesextractsusingUHPLC–HRMS.J.FoodEng.,

http://dx.doi.org/10.1016/j.jfoodeng.2014.10.021.

Chemat,F.,Vian,M.A.,Cravotto,G.,2012.Greenextractionofnaturalproducts: conceptandprinciples.Int.J.Mol.Sci.13,8615–8627.

Cheok,C.Y.,Chin,N.L.,Yusof,Y.A.,Talib,R.A.,Law,C.L.,2012.Optimizationoftotal phenoliccontentextractedfromGarciniamangostanaLinnhullusingresponse surfacemethodologyversusartiﬁcialneuralnetwork.Ind.CropsProd.40, 247–253.

Dahmoune,F.,Boulekbache,L.,Moussi,K.,Aoun,O.,Spigno,G.,Madani,K.,2013.

ValorizationofCitruslimonresiduesfortherecoveryofantioxidants: evaluationandoptimizationofmicrowaveandultrasoundapplicationto solventextraction.Ind.CropsProd.50,77–87.

Dahmoune,F.,Moussi,K.,Remini,H.,Belbahi,A.,Aoun,O.,Spigno,G.,Madani,K., 2014.Optimizationofultrasound-assistedextractionofphenoliccompounds fromCitrussinensisLpeelsusingresponsesurfacemethodology.Chem.Eng. 37,889–894.

Dahmoune,F.,Nayak,B.,Moussi,K.,Remini,H.,Madani,K.,2015.Optimizationof microwave-assistedextractionofpolyphenolsfromMyrtuscommunisL.leaves. FoodChem.166,585–595.

Dawidowicz,A.L.,Olszowy,M.,Jó ´zwik-Dol ˛eba,M.,2015.Antagonisticantioxidant effectinbutylatedhydroxytoluene/butylatedhydroxyanisolemixture.J.Food Process.Preserv.,http://dx.doi.org/10.1111/jfpp.12469(Inpress).

Delpino-Rius,A.,Eras,J.,Vilaró,F.,Cubero,M.Á.,Balcells,M.,Canela-Garayoa,R., 2015.Characterisationofphenoliccompoundsinprocessedﬁbresfromthe juiceindustry.FoodChem.172,575–584.

Geyikçi,F.,Kılıç,E.,Çoruh,S.,Elevli,S.,2012.Modellingofleadadsorptionfrom industrialsludgeleachateonredmudbyusingRSMandANN.Chem.Eng.J. 183,53–59.

Ghoreishi,S.M.,Heidari,E.,2013.ExtractionofEpigallocatechin-3-gallatefrom greenteaviasupercriticalﬂuidtechnology:neuralnetworkmodelingand responsesurfaceoptimization.J.Supercrit.Fluids74,128–136.

Gong,Y.,Hou,Z.,Gao,Y.,Xue,Y.,Liu,X.,Liu,G.,2012.Optimizationofextraction parametersofbioactivecomponentsfromdefattedmarigold(TageteserectaL.) residueusingresponsesurfacemethodology.FoodBioprod.Process.90,9–16.

Gruenwald,J.,2009.Novelbotanicalingredientsforbeverages.Clin.Dermatol.27, 210–216.

Haﬁzi,A.,Ahmadpour,A.,Koolivand-Salooki,M.,Heravi,M.,Bamoharram,F., 2013a.ComparisonofRSMandANNfortheinvestigationoflinear

alkylbenzenesynthesisoverH14[NaP5W30O110]/SiO2catalyst.J.Ind.Eng. Chem.19,1981–1989.

Haﬁzi,A.,Ahmadpour,A.,Koolivand-Salooki,M.,Heravi,M.M.,Bamoharram,F.F., 2013b.ComparisonofRSMandANNfortheinvestigationoflinear alkylbenzenesynthesisoverH14[NaP5W30O110]/SiO2catalyst.J.Ind.Eng. Chem.19,1981–1989.

Huang,D.,Ou,B.,Hampsch-Woodill,M.,Flanagan,J.A.,Prior,R.L.,2002.

High-throughputassayofoxygenradicalabsorbancecapacity(ORAC)usinga multichannelliquidhandlingsystemcoupledwithamicroplateﬂuorescence readerin96-wellformat.J.Agric.FoodChem.50,4437–4444.

Jaramillo-Flores,M.E.,González-Cruz,L.,Cornejo-Mazón,M.,Dorantes-Alvarez,L., Gutiérrez-López,G.F.,Hernández-Sánchez,H.,2003.Effectofthermal treatmentontheantioxidantactivityandcontentofcarotenoidsandphenolic compoundsofcactuspearcladodes(Opuntiaﬁcus-indica).FoodSci.Technol. Int.9,271–278.

Jerman,T.,Trebˇse,P.,MozetiˇcVodopivec,B.,2010.Ultrasound-assistedsolidliquid extraction(USLE)ofolivefruit(Oleaeuropaea)phenoliccompounds.Food Chem.123,175–182.

Kadam,S.U.,Tiwari,B.K.,Smyth,T.J.,O’Donnell,C.P.,2015.Optimizationof ultrasoundassistedextractionofbioactivecomponentsfrombrownseaweed Ascophyllumnodosumusingresponsesurfacemethodology.Ultrason. Sonochem.23,308–316.

Kumar,R.,Chauhan,S.,2015.Studyonsurfaceroughnessmeasurementforturning ofAl7075/10/SiCpandAl7075hybridcompositesbyusingresponsesurface methodology(RSM)andartiﬁcialneuralnetworking(ANN).Measurement65, 166–180.

Lupien,J.R.,2002.Implicationsforfoodregulationsofnovelfood:Safetyand labeling.AsiaPac.J.Clin.Nutr.11,S224–S229.

Machado,A.P.D.F.,Pasquel-Reátegui,J.L.,Barbero,G.F.,Martínez,J.,2015. Pressurizedliquidextractionofbioactivecompoundsfromblackberry(rubus fruticosusl.)residues.acomparisonwithconventionalmethods.FoodRes.Int.,

http://dx.doi.org/10.1016/j.foodres.2014.12.042(Inpress).

Molyneux,P.,2004.Theuseofthestablefreeradicaldiphenylpicrylhydrazyl (DPPH)forestimatingantioxidantactivity.SongklanakarinJ.Sci.Technol.26, 211–219.

Pilkington,J.L.,Preston,C.,Gomes,R.L.,2014.Comparisonofresponsesurface methodology(RSM)andartiﬁcialneuralnetworks(ANN)towardsefﬁcient extractionofartemisininfromArtemisiaannua.Ind.CropsProd.58,15–24.

Raﬁgh,S.M.,Yazdi,A.V.,Vossoughi,M.,Safekordi,A.A.,Ardjmand,M.,2014.

Optimizationofculturemediumandmodelingofcurdlanproductionfrom PaenibacilluspolymyxabyRSMandANN.Int.J.Biol.Macromol.70,463–473.

Rami ć,M.,Vidovi ć,S.,Zekovi ć,Z.,Vladi ć,J.,Cvejin,A.,Pavli ć,B.,2015.Modeling andoptimizationofultrasound-assistedextractionofpolyphenolic compoundsfromAroniamelanocarpaby-productsfromfilter-teafactory. Ultrason.Sonochem.23,360–368.

Rodrigues,S.,Fernandes,F.A.N.,deBrito,E.S.,Sousa,A.D.,Narain,N.,2015.

Ultrasoundextractionofphenolicsandanthocyaninsfromjabuticabapeel.Ind. CropsProd.69,400–407.

Sarve,A.,Sonawane,S.S.,Varma,M.N.,2015.Ultrasoundassistedbiodiesel productionfromsesame(SesamumindicumL.)oilusingbariumhydroxideasa heterogeneouscatalyst:comparativeassessmentofpredictionabilities betweenresponsesurfacemethodology(RSM)andartiﬁcialneuralnetwork (ANN).Ultrason.Sonochem.26,218–228.

Shojaeimehr,T.,Rahimpour,F.,Khadivi,M.A.,Sadeghi,M.,2014.Amodelingstudy byresponsesurfacemethodology(RSM)andartiﬁcialneuralnetwork(ANN) onCu2+adsorptionoptimizationusinglightexpendedclayaggregate(LECA).J. Ind.Eng.Chem.20,870–880.

Spigno,G.,Tramelli,L.,DeFaveri,D.M.,2007.Effectsofextractiontime, temperatureandsolventonconcentrationandantioxidantactivityofgrape marcphenolics.J.FoodEng.81,200–208.

Thabit,R.A.,Cheng,X.-R.,A.L-Hajj,N.,Rahman,M.R.T.,Le,G.,2015.Antioxidantand Conyzabonariensis:aReview.Eur.Acad.Res.2(6),8454–8474.

Vázquez,C.V.,Rojas,M.G.V.,Ramírez,C.A.,Chávez-Servín,J.L.,García-Gasca,T., FerrizMartínez,R.A.,García,O.P.,Rosado,J.L.,López-Sabater,C.M.,Castellote, A.I.,Montemayor,H.M.A.,delaTorreCarbot,K.,2015.Totalphenolic compoundsinmilkfromdifferentspeciesdesignofanextractiontechniquefor quantiﬁcationusingtheFolin–Ciocalteumethod.FoodChem.176,480–486.

Veggi,P.C.,Prado,J.M.,Bataglion,G.A.,Eberlin,M.N.,Meireles,M.A.A.,2014.

Obtainingphenoliccompoundsfromjatoba(HymenaeacourbarilL.)barkby supercriticalﬂuidextraction.J.Supercrit.Fluids89,68–77.