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 READS155
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
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
aaLaboratoireBiomathématiquesBiophysiqueBiochimieetdeScientométrie(L3BS),FacultédesSciencesdelaNatureetdelaVieUniversitédeBejaia,06000
Bejaia,Algeria
bFacultédesSciencesdelaNatureetdelaVieetdesSciencesdelaTerre,UniversitédeBouira,10000Bouira,Algeria
cLaboratoireBiomathématiquesBiophysiqueBiochimieetdeScientométrie(L3BS),FacultédesSciencesdelaNatureetdelaVieetdesSciencesdelaTerre,
UniversitédeKhemisMiliana,44225KhemisMiliana,Algeria
dTechnicalSciencesDepartment,InstituteofSciences,UniversityofAkliMohandOulhadj,Bouira,Algeria eSchoolofFoodandAgriculture,UniversityofMaine,Orono,ME04469,UnitedStates
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.
©2015ElsevierB.V.Allrightsreserved.
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
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)fullfactorialcentralcompositedesign(CCD)inRSMand
anANNbasedmodelweredevelopedand comparedtoindicate howthepercentageofTPCrecovery,throughthegreenextraction processofP.lentiscusleaves,mightbeoptimized.The experimen-talvariablesincludingethanolproportion,amplitudeintensityand extractionduration,inanextractionprocedurewhichisconsidered toapproximatetheindustrialapplication.Thedevelopedmodels werethencomparedintheirsuitabilityforpredictingTPCrecovery byanalysingtheircoefficientofdetermination(R2),adjusted
coef-ficientofdetermination(R2adj),rootmeansquareerror(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,thepowderwaspassedthroughstandard125msieveand storedinairtightbagsunderdarknessuntiluse.Moisturecontent wasassessedbyconstantweightat105◦Candwas5.0±0.5%,the wateractivityAw(HygroPalmAw)was0.18±0.2at20.6◦C.
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 (0.45m)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,100Loftheextractwasmixedwith750L Folin–Ciocalteudilutedreagent(1/10).Then, thesolutions were mixedandleftatroomatroomtemperaturefor5min.Afterthat, 750Lof7.5%sodiumcarbonate(Na2CO3)solutionwasadded.
Afterincubationat25◦Cfor90mintheabsorbancewasmeasured at725nm(1cmopticalpath)againstablank(madeasreportedfor thesamplebutwith100Lofsamplesolvent)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
mea-suredonaplatereader(OmegaFLUOstar,BMGLabtech,Cary,NC, USA).300LofDPPH•solutioninmethanol(70M)wasmixed with10Lofextractandthemixturewasincubatedfor20min at37◦C.Thedecreaseinabsorbancereadingofthemixturewas measuredat515nm.Theantioxidantcapacityoftheextractwas expressedasapercentageofinhibitionofDPPHradical(% inhibi-tionofDPPHradical)calculatedaccordingtothefollowingequation (Eq.(2)):
%inhibition=AOX=Ablankt−20Asamplet−20 ADPPHt−0
(2) where,Ablankistheabsorbancevalueoftheblank(300LofDPPH
solutionplus10Lofthesolventinwhichtheextracthasbeen 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 of20Lwithsuitabledilutioninphosphatebuffersalinesolution (75mM,pH7.0)wereloadedtopolystyrene96-wellmicroplatesin triplicatebasedonarandomisedsetlayout.
Thesamplesweredilutedtotheproperconcentrationrangefor fittingthelinearityrangeofthestandardcurve.Afterloading20L ofsample,standardandblank,and200Lofthefluorescein solu-tionintoappointedwellsaccordingtothelayout,themicroplate (sealedwithfilm)wasincubatedfor15minintheplatereaderat 37◦Candthen20Lofperoxylgenerator2,2 -azobis(2-amidino-propanedihydrochloride(AAPH)(3.2M)wasaddedtoinitiatethe oxidationreaction.Theplatereaderwasprogrammedtoinjectand recordthefluorescenceoffluoresceinoneverycycle.Thekinetic readingdecreaseofFluoresceinintensity(%)wasrecordedfor60 cycleswith40spercyclesettingduring40minat37◦C.Absorbance readingsoftheplateweretakeneverycycleusinganexcitation wavelengthof485nmandanemissionwavelengthof535nmuntil allfluorescencereadingsdeclinedtoless than25%oftheinitial values.
FourTroloxsolutions(6.25,12.5,25,50Minphosphatebuffer 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-trationswasdeterminedandORACvalueswereexpressedasmol 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),adjustedcoefficient(R2adj),
therootmean-squar-error (RMSE),adequationprecesion (Adeq prec)ofthefunctionwereusedforevaluatingthestatistical sig-nificanceofthemodeldeveloped(Table2).
Table2
Analysisofvariance(ANOVA)fortheexperimentalresultsobtainedbyusing ultra-soundassistedextraction.
Source Sumofsquares DF*** F-value P-valueProb>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 R2adj 0.96 Adeqpre* 26.25 RMSE** 4.65 CorTotal 9165.69 18 *Adequationprecision. **Rootmeansquareerror. ***Degreeoffreedom.
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),therootmean-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.
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 R2Adj=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=
Ni=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(R2adj),rootmeansquareerror (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(R2Adj)arecloseto1,whichrevealed
also,thatthereareexcellentcorrelationsbetweenthepredicted andexperimentalmodels(Fig.2a).Inaddition,thesmallvaluesof CV(3.45%)than10andhighervaluesofadequateprecisionthan4 forbothextractionplantsgivebetterreproducibilityandadequate signal(Table2).Moreover,“The ¨‘‘PredR-Squared” ¨of0.8422isin reasonableagreementwiththe ¨‘‘AdjR-Squared” ¨of0.9630;¨i.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.
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
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.
-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 demonstratesa superioraccuracy
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).
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.
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)andANN(R2=0.999)areusefuland
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.
Geyikc¸i,F.,Kılıc¸,E.,C¸oruh,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 ´c,M.,Vidovi ´c,S.,Zekovi ´c,Z.,Vladi ´c,J.,Cvejin,A.,Pavli ´c,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.