MathematicalMedicineandBiologyPage1of 26 doi:10.1093/imammb/dqnxxx 4 Delayeffects inthe response oflowgrade gliomasto radiotherapy: A 1 0 mathematical model andits therapeutical implications 2 n V´ICTOR M.PE´REZ-GARC´IA∗ a J DepartamentodeMatema´ticas,UniversidaddeCastilla-LaMancha ETSIIndustriales,Avda.CamiloJose´ Cela3,13071CiudadReal,Spain. 2 1 ∗Correspondingauthor:[email protected] ] MAGDALENA BOGDANSKA M MathematicsDepartment,TechnicalUniversityofLodz, Q Lodz,Wolczanska214Street,Poland. o. ALICIA MART´INEZ-GONZA´LEZ,JUAN BELMONTE-BEITIA i DepartamentodeMatema´ticas,UniversidaddeCastilla-LaMancha b ETSIIndustriales,Avda.CamiloJose´ Cela3,13071CiudadReal,Spain. - q [ PHILIPPESCHUCHT Universita¨tsklinikfu¨rNeurochirurgie 1 v BernUniversityHospital,CH-3010Bern,Switzerland 3 LUIS A.PE´REZ-ROMASANTA 0 6 RadiotherapyUnit,UniversityHospitalofSalamanca,Salamanca,Spain 2 . [ReceivedonXXXXX] 1 0 Lowgradegliomas(LGGs)areagroup ofprimarybraintumorsusuallyencountered inyoung patient 4 populations.Thesetumorsrepresentadifficultchallengebecausemanypatientssurviveadecadeormore 1 andmaybeatahigherriskfortreatment-relatedcomplications. Specifically,radiationtherapyisknown : v to have a relevant effect on survival but in many cases it can be deferred to avoid side effects while i maintaining its beneficial effect. However, a subset of low-grade gliomas manifests more aggressive X clinicalbehaviorandrequiresearlierintervention. Moreover,theeffectivenessofradiotherapydepends r onthetumorcharacteristics.RecentlyPalludetal.,[Neuro-oncology,14(4):1-10,2012],studiedpatients a with LGGstreated with radiation therapy as a first line therapy. and found the counterintuitive result thattumorswithafastresponsetothetherapyhadaworseprognosisthanthoserespondinglate. Inthis paperweconstructamathematicalmodeldescribingthebasicfactsofgliomaprogressionandresponseto radiotherapy.ThemodelprovidesalsoanexplanationtotheobservationsofPalludetal.Usingthemodel weproposeradiationfractionationschemesthatmightbetherapeuticallyuseful byhelpingtoevaluate thetumormalignancywhileatthesametimereducingthetoxicityassociatedtothetreatment. Keywords:LowGradeGliomas,Radiotherapy,Mathematicalmodeloftumorresponse 1. Introduction Lowgradeglioma(LGG)isatermusedtodescribeWHOgradeIIprimarybraintumorsofastrocytic and/or oligodendroglialorigin. These tumors are highly infiltrative and generally incurable but have mediansurvivaltimeof>5yearsbecauseoflowproliferation(Pignattietal.,2002;Pouratian&Schiff, 2010). Whilemostpatientsremainclinicallyasymptomaticbesidesseizures,thetumortransformation (cid:13)c Theauthor2008.PublishedbyOxfordUniversityPressonbehalfoftheInstituteofMathematicsanditsApplications.Allrightsreserved. 2of26 V.M.PE´REZ-GARC´IAetal. toaggressivehighgradegliomaiseventuallyseeninmostpatients. ManagementofLGGhashistoricallybeencontroversialbecausethesepatientsaretypicallyyoung, with few, if any, neurologicalsymptoms. Historically, a wait and see approach was often favored in mostcases of LGG due to the lack of symptomsin these mostly youngand otherwisehealthyadults. Thesupportforthispracticecamefromseveralretrospectivestudiesshowingnodifferenceinoutcome (survival, qualityof life) if therapywas deferred(Olsonetal., 2000;Grier&Batchelor, 2006). Other investigationshavesuggestedaprolongedsurvivalthroughsurgery(Smithetal.,2008). Inabsenceof arandomizedcontrolledtrial,recentlypublishedstudiesmayprovidethemostconvincingevidencein supportofanearlysurgerystrategy(Jakolaetal.,2012) andwaitingforthe useofothertherapeutical options such as radiotherapy and chemotherapy. However, the decision on the individual treatment strategy is based on a number of factors including patient preference, age, performance status, and locationoftumor(Ruiz&Lesser,2009;Pouratian&Schiff,2010). As to radiation therapy the clinical trial by Garciaetal (2004) showed the advantageof using ra- diotherapy in addition to surgery. However, the timing of radiotherapy after biopsy or debulking is debated. Itisnowwellknownthatimmediateradiotherapyaftersurgeryincreasesthetimeofresponse (progression-free survival), but does not seem to improve overal survival while at the same time in- ducingseriousneurologicaldeficitsasa resulttonormalbraindamage(VandenBent,2005). Overall survivaldependsmoreonpatientandtumour-relatedcharacteristicssuchasage,tumourgrade,histol- ogy and neurologicalfunction than the details of the plan of radiotherapytreatment. Radioteraphyis usuallyofferedforpatientswithacombinationofpoorriskfactorssuchasage,sub-totalresection,and diffuseastrocytomapathology(Higuchietal.,2004). Mathematicalmodellinghasthepotentialtoselectpatientsthatmaybenefitfromearlyradiotherapy. Also it may help in developingspecific optimalfractionationschemesfor selected patientsubgroups. Howeverdespiteitsenormouspotential,mathematicalmodellinghashadalimitedusewithstrongfocus onsomeaspectsofradiationtherapy(RT)forhigh-gradegliomas(Rockneetal.,2010;Bondiauetal., 2011;Konukogluetal.,2010;Barazzuoletal.,2010). Uptonow,noideascomingfrommathematical modellinghavebeenfoundusefulforclinicalapplication. Thereisthusaneedformodelsaccountingforthefundamentalfeaturesoflow-gradegliomadynam- icsandtheirresponseto radiationtherapywithoutinvolvingexcessivedetailsonthe -oftenunknown- specificprocessesbutallowingthequalitativeunderstandingofthephenomenainvolved.Theavailabil- ityofsystematicandquantitativemeasurementsofLGGgrowthratesprovideskeyinformationforthe developmentandvalidationofsuchmodels(Palludetal.,2012,b). Inthispaperwepresentasimplemathematicalmodelcapturingthekeyfeaturesseenintheresponse of LGGsto radiation. Our modelincorporatesthebasic elementsoftumordynamics: infiltrationand invasionofthenormalbrainbythetumorcells,proliferationandtumorcelldeathinresponsetother- apy. Radiationtherapyis includedin analmostparameter-freeway thatcapturesthe essentials of the dynamicsandexplainstherelationshipbetweenproliferation,responsetothetherapyandprognosisas recentlyreportedbyPalludetal.(2012b). In addition to explaining the counterintuitive observationsof Palludetal. (2012b) the model pre- sentedinthispapercanbeusedtoexploredifferentradiationregimes. Theanalysistobepresentedin thispapersuggeststhepossibilityofusingradiationtherapywithpalliativeintentandalsototestwhat thetumorresponseisandhelptheoncologistsinmakingthebestpossibledecisionsonwhenandhow toactonthetumor. Ourplaninthispaperisasfollows. First,inSection2wepresentourmodelaccountingfortumor celldynamicsandtheresponseofthetumorcellstoradiation. NextinSec. 3wepresenttheresultsof thenumericalsimulationsofthemodelindifferentscenariosandstudythedependenceofthemodelon DELAYEFFECTSINRADIOTHERAPYOFLGG 3of26 thedifferentparameters. InSec. 4wedisplaysomeanalyticalestimatesofthetypicaldynamicsofthe tumorresponsetoradiation. InSec. 5wediscusssomehypothesissuggestedfromthemodelthatmay beusefulfortherapy.Finally,inSec. 6wesummarizeourconclusions. 2. MathematicalmodelfortheresponseofLGGstoradiotherapy 2.1 Tumorcellcompartment Inthelastyearstherehasbeenalotofactivityonmathematicalmodelsofgliomaprogression(Murray, 2007;Stamatakosetal.,2006,b;Frieboesetal.,2007;Swansonetal.,2008;Bondiauetal.,2008;Eikenberryetal., 2009;Tanakaetal.,2009;Konukogluetal.,2010;Wangetal.,2009;Rockneetal.,2010;Pe´rez-Garc´ıaetal., 2011;Pe´rez-Garc´ıa&Mart´ınez-Gonza´lez,2012;Hatzkirouetal.,2012;Badoualetal.,2012;Mart´ınez-Gonza´lezetal., 2012;Guetal.,2012;PainterandHillen,2013).Inthispaperwewillconsideramodelforthecompart- mentof tumorcells of the simplest possible type: a Fisher-Kolmogorov(FK) type equation(Murray, 2007). Morecomplicatedmodelssuchassingle-cellbasedmodelswouldallow,inprinciple,tofollow theindividualmovementofthetransformedastrocytesthroughthebrainparenchyma.However,consid- eringthatthebasicrulesbehindamodelaremoreimportantthanthemodeldetails,wehavediscarded boththeuseofon-latticemodels,whicharenotrealisticwhencellmotionisconsidered,andoff-lattice modelswhichassumefixedcellgeometriesand/orincorporateunknowncell-cellinteractions. Besides, thesemodelsoftenrequiretheestimationofalargenumberofunknownparametersandthedetermina- tionofinitialcellconfigurations,whichareextremelydifficulttovalidateininvivoexperimentsand/or usingclinicaldata. Thus,tokeepourdescriptionassimpleasposiblewehaveoptedforacontinuous modelasfollows ¶ C = DD C+r (1 C)C, (2.1a) ¶ t − C(x,0) = C (x), (2.1b) 0 whereC(x,t) is the tumor cell density as a function of time t and the spatial position x and it is measuredinunitsofthemaximalcelldensityallowedinthe tissueC (typicallyaround103 cell/cm). D =(cid:229) n ¶ 2/¶ x2 isthen dimensionalLaplacianoperator. ∗ j=1 j − D is thediffusioncoefficientaccountingfortheaveragecellularmotilitymeasuredin mm2/dayto beassumedinthispapertobeconstantandspatiallyuniform.Migrationingliomasisnotsimpleandin factmanyauthorshaveproposedthatthehighlyinfiltrativenatureofhumangliomasrecapitulatesthe migratorybehaviorofglialprogenitors(Suzukietal.,2002;Dirks,2001). Hereweassume,asinmost models,thatgliomacellinvasionthroughoutthebrainisbasicallygovernedbyastandardFickiandiffu- sionprocess.Morerealisticandcomplicateddiffusiontermsingliomasshouldprobablybegovernedby fractional(anomalous)diffusion(Fedotovetal.,2011)orothermoreelaborateterms(Deroulersetal., 2009) to accountfor the high infiltration observedin this type of tumors(Onishietal., 2011) and the fact that cells do not behave like purely random walkers and may actually remain immobile for long timesbeforecompelledtomigratetoamorefavorableplace. Inadditioninrealbraintherearespatial inhomogeneities expected in the parameter values such as different propagation speeds in white and graymatter,andanisotropies(e.g.onthediffusiontensorwithpreferentialpropagationdirectionsalong white matter tracts). Many papers have incorporated these details (Jbadietal., 2005; Bondiauetal., 2008;Konukogluetal.,2010;Clatz,2005;PainterandHillen,2013)mostlywiththeintentiontomake patient-specificprogressionpredictions. However,themainlimitationisthelackofinformationonthe (many) patientspecific unknowndetails what has limited progressesin that direction. Thus, in order 4of26 V.M.PE´REZ-GARC´IAetal. to simplify the analysis and focus on the essentials of the phenomena we have chosen to study the modelinonespatialdimensionandinisotropicmedia.Itisinterestingthatuptonowonlythesimplest models such as those given by Eqs. (2.1) have been used to extract conclusionsuseful for clinicians (Swansonetal.,2008;Wangetal.,2009). The choiceof one-dimensionaldiffusionintendsto incorporatequalitativelydiffusionphenomena inthesimplestpossibleway. Frontsolutionsofthe1DFKequationhavebeenextensivelystudiedand areknowntopropagatewithaminimalspeedc=2√r Dwhenstartingfromstillinitialdata(Murray, 2007). It is interesting that the 1D model recapitulates the most relevant -for us-phenomenologyob- servedinhigher-dimensionalscenarios. Firstitisobviousandwell-knownthatfront(invasionwaves) solutions of the 1D FK equation also solve higher dimensional version of the equation (Xin, 2000; BrazhnikandTyson,2000). Moreover,thosesolutionsareasymptoticallystable(Xin,2000;Sattinger, 1976)whatmeansthatunlikeothermorecomplicatednon-symmetricalsolutions(BrazhnikandTyson, 2000), they do arise as limits of non symmetric initial data. It is also well known that radially sym- metric (in 2D) or spherically symmetric (in 3D) travelling wave solutions of FK do not exist in high dimensions but that symmetric fronts also develop in those scenarios with a non-constant speed that dependsonthelocalcurvatureR(BrazhnikandTyson,2000;Volpert&Petrovskii,2009). Asthefront growswith time the now radius-dependentfrontspeed is givenby c(R)=c D/R(t). Thusgrowing symmetricmultidimensionalsolutionswithlargecurvatureradii(R ¥ )gro−wwiththesamespeedas → 1Dfronts(Volpert&Petrovskii,2009;GerleeandNelander,2012). In this paper we are interested on the descriptionof low-gradegliomasthat typically are very ex- tended when diagnosed, thus the initial data radius is large and fronts are well developed by then. Althoughduring the initial stages of tumordevelopmentthe dimensionalitymay play a relevantrole, for spatially extended tumors the effect of using higher-dimensional operators is not expected to be substantial. Moreover, some of the phenomena to be described later in this paper are found to be essentially independentof diffusionand a verygood qualitativeagreementwill be foundbetween our simplified analysisandthegrowthdynamicsofthemeantumordiameter. Takingintoaccountalltheseevidences we will keep the system one-dimensional,since our intention is notto providea detailed quantitative descriptionoftheprocesses-thatinanywaywouldbebeyondthereachofasimplemodelsuchasFK- but instead to provide a qualitative description of the dynamics in the simplest possible way. As we willdiscussindetaillater,thisapproachwillleadtoasimpleyetqualitativelycorrectdescriptionofthe responseoflowgradegliomastoradiotherapy. The parameterr in Eq. (2.1) is the proliferationrate (1/day)its inverse givingan estimate of the typicalcelldoublingtimes. Wehavechosenalogistictypeofproliferationleadingtoamaximumcell densityC(x,t)=1. Finally,thetumorevolvesfromaninitialcelldensitygivenbythefunctionC (x) 0 inanunboundeddomain,soweimplicitlyassumeittobelocatedinitiallysufficientlyfarfromthegrey matter. A very interesting feature of model Eqs. (2.1) is the well known fact that a tumor front arises propagatingattheasymptotic(constant)speedofv =2√Dr whatisinverygoodagreementwiththe ∗ observed fact that the tumor mean diameter grows at an approximately constant speed (Palludetal., 2012). Whilemanyothermathematicalmodelsofgliomasincorporatedifferentcellphenotypes,e.g. nor- moxic(proliferative)andhypoxic(migratory)phenotypes,suchasinMart´ınez-Gonza´lezetal.(2012); Hatzkirouetal.(2012);Pe´rez-Garc´ıa&Mart´ınez-Gonza´lez(2012),herewefocusourattentiononLGGs andassuchwillconsiderasingle(dominant)tumorcellphenotype.Inourmodelwedonotincludethe possibleexistenceofdifferenttumorcellpopulationswithdifferentsensitivitiestotherapysuchasstem DELAYEFFECTSINRADIOTHERAPYOFLGG 5of26 cellssincethefunctionandmechanismsofstemcellsinglioblastomaareyetunderdebate(Barrettetal., 2012;Chenetal.,2012). 2.2 Responsetoradiation Radiationtherapyhasbeenincorporatedindifferentformstomathematicalmodelsofhigh-gradeglioma progression(Rockneetal.,2010;Bondiauetal.,2011;Konukogluetal.,2010;Barazzuoletal.,2010). In thispaperwe wantto focusourattentiononLGGswhoseresponsetoradiationis verydifferentto the oneobservedinHGGs. RadiationtherapyinLGGstypicallyinducesa responsethatprolongsfor severalyearsaftertherapy. Very interesting quantitative data on the response of LGGs to radiation have been reported in a retrospectivestudybyPalludetal.(2012). TheauthorsstudiedpatientsdiagnosedwithgradeIILGGs treatedwithfirst-lineradiotherapybeforeevidencesofmalignanttransformation. Patientswithapost- RTvelocityofdiametricexpansion(VDE)(Palludetal.,2012)slowerthan-10mm/yearweretakenasa subgroupofslowlygrowingLGGs.Patientswithapost-RTVDEof-10mm/yearorfasterwereincluded inthegroupoffastgrowingLGGs. Theauthorsconcludedthatthepost-RTVDEwassignificantlyfaster inthegroupwithhighproliferation.Alsointhepatientswithanavailablepre-RTVDE,thelowpre-RT VDEsubgrouppresentedaslowerVDEatimagingprogression. Astothesurvivaltime(ST),thepost- radiotherapyVDEcarriedaprognosticsignificanceonsurvivaltime,asthefastpost-radiotherapytumor volumedecrease(VDE at-10 mm/yearor faster)were associated with a significantlyshortersurvival thantheslowpost-radiotherapytumorvolumedecrease(VDEslowerthan-10mm/year). Theveryslowresponsetoradiotherapy,leadingtotumorregressionlastingforseveralyearsisdiffi- culttounderstandinthecontextofthestandardlinearquadraticmodelinwhichdamageisinstantaneous andleadstocelldeathearlyafterradiationtherapy. Howeverakeyaspectofthecellresponsetoradia- tionisthatirradiatedcellsusuallydiebecauseoftheso-calledmitoticcatastropheaftercompletingone orseveralmitoses(VanderKogel&Joiner,2009). Thismeansthatslowlyproliferatingtumors,asitis the caseofLGGswithtypicallow proliferationindexesbetween1%and5%inpathologicalanalyses needaverylongtimetomanifesttheaccumulatedcelldamagethatcannotberepaired. Thus,inordertocaptureinaminimalwaytheresponseofthetumortoradiationwewillcomplement equation(2.1)forthedensityoffunctionallyalivetumorcellsC(x,t)withanequationfortheevolution of the density of irreversibly damaged cells after irradiationC (x,t). Our model for the evolution of d bothtumorcelldensitieswillbegivenbytheequations ¶ C = DD C+r (1 C C )C, (2.2a) ¶ t − − d ¶ C r d = DD C (1 C C )C . (2.2b) ¶ t d− k − − d d ThefirstequationisaFisher-KolomogorovtypeequationdescribingtheevolutionoftumorcellsC(x,t). Thesaturationtermincludesthetotaltumorcelldensity,i.e. boththefunctionaltumorcellsandthose damaged by radiationC (x,t). The evolution of cells irreversibly damaged by radiation is given by d Eq. (2.2b). Asitiswelldescribedintheliterature(VanderKogel&Joiner,2009),mostofthesecells behavenormallyuntilacertainnumberofmitosiscycles,thuswewillconsiderthatafteranaverageof k mitosis cyclesthese cells die resultingin a negativeproliferation. The longersurvivaltime kt with t =1/r beingthetumorpopulationdoublingtime,resultsinareducedproliferationpotentialr /k,that is the coefficientusedforthe negativeproliferationterm. Thus, theparameterk in Eq. (2.2b) hasthe meaningoftheaveragenumberofmitosiscyclesthatdamagedcellsareabletocompletebeforedying. 6of26 V.M.PE´REZ-GARC´IAetal. Aswiththenormalpopulationthenumberofcellsenteringmitosisdependsinanonlinearwayonboth tumorcellpopulations(cf. lastterminEq.(2.2b)). Wewillassumeaseriesofradiationdosesd ,d ,...,d givenattimest ,t ,...,t . Theinitialdatafor 1 2 n 1 2 n thefirstsubintervalwillbegivenbytheequations C(x,t ) = C (x), (2.3a) 0 0 C (x,t ) = 0. (2.3b) d 0 The evolution of the tumor follows then Eqs. (2.2) until the first radiation dose d , given at time t . 1 1 Theirradiationresultsinafractionofthecells(survivingfraction)abletorepairtheradiation-induced damagegivenbyS (d )andafraction1 S (d )ofcellsunabletorepairtheaccumulateddamagethus f 1 f 1 − feedingthecompartmentofirreversiblydamagedcells. Thesubsequentevolutionofthepopulationsis againgivenbyEqs. (2.2)untilthenextRTdoseisgiven.Ingeneralaftereachirradiationeventweget C(x,t+) = S (d )C(x,t ), (2.3c) j f j −j Cd(x,t+j ) = Cd(x,t−j )+ 1−Sf(dj) C(x,t−j ), (2.3d) whereS (d )isthesurvivalfractionafteradoseofradi(cid:2)ationd ,i.e(cid:3).thefractionofdamagedtumorcells f j j after irradiation that are not able to repair lethal damage and are incorporatedto the compartmentof damagedcells. Forthedosestobeconsideredindependenttheintervalbetweendoses(typically1day) hastobelargerthanthetypicaldamagerepairtimes(oftheorderofhours). Theevolutionofbothcell densitiesbetweenirradiationeventsisgivenbythePDEs(2.2). 2.3 Parameterestimation Eqs.(2.2)togetherwiththeinitialconditionsforeachsubproblem(2.3)definecompletelythedynamics ofalowgradegliomaintheframeworkofoursimplifiedtheoreticalapproach. WehavechosentheparameterstodescribethetypicalgrowthpatternsofLGG.Fortheproliferation rate we have chosen typical values to be small and around r =0.003 day 1 (see e.g. Badoualetal. − (2012)),thatgivedoublingtimesoftheorderofayear. Specificallywehaveconsideredvaluesranging fromr =0.001day 1 forveryslowlygrowingLGGstor =0.01day 1. Forthecelldiffusioncoef- − − ficient we have taken values around D=0.0075mm2 /day (Jbadietal., 2005). This choice, together withthepreviouslychosenr leadstoasymptotictumordiametergrowthspeedsgivenbyv=4√Dr of the orderofseveralmillimetersperyear, inagreementwith typicaldiametricgrowthspeedsofLGGs (Palludetal., 2012). Howeverthe factthat the asymptotic speed is only reachedwhen the tumorcell densityisaroundonemayrequiretakinglargervaluesofDtomatchtherealgrowthspeeds. Astotheradiobiologicalparameters,beinggliomasveryresistanttoradiationwehavetakenvalues intherangeS (1.8Gy) SF 0.9consideringthemediansurvivalfractionvalue0.83afteronedose f 1.8 ≡ ∼ of2GygivenbyBarazzuoletal.(2010). Finally, the average number of mitosis cycles completed before the mitotic catastrophe occurs is difficult to estimate. This parameter intends to summarize in a single number a complex process in which a cell hit by radiation and its progeny die after some more mitosis cycles leading to a final extinctionafteravariabletime. Deathbymitoticcatastropheimpliesaminimalvalueofk=1andto allowforsomemoretimewemaychoosevaluesintherangek=1 3(VanderKogel&Joiner,2009). − Wewillshowlaterthatthechoiceofthisparameterhasalimitedeffectonthemodeldynamicsandthat instandardfractionationschemestheremaybebiologicalreasonstotakeitask=1. OurtypicalchoicesforthefullsetofmodelparametersissummarizedinTable1. DELAYEFFECTSINRADIOTHERAPYOFLGG 7of26 Variable Description Value(Units) References C Maximumtumor 106cell/cm2 Swansonetal.(2008) ∗ celldensity D Diffusioncoefficientfor 0.01mm2/day Jbadietal.(2005) tumorcells r Proliferation 0.00356day 1 Badoualetal.(2012) − rate SF Survivalfraction 0.9 Barazzuoletal.(2010) 1.8 ∼ for1.8Gy k Averagenumberofmitosis 1-3 VanderKogel&Joiner(2009) cyclescompletedbeforethe mitoticcatastrophe Table1.TypicalvaluesofthebiologicalparametersinthemodelofLGGevolution. Inthispaperwewillfixthedoseperfractioninagreementwiththestandardfractionationschemes forLGGstobe1.8Gy,theonlyrelevantparameteristhesurvivalfractionSF thatwillbetakentobe 1.8 aroundSF 0.9,asdiscussedabove. Inmanyexampleswewillchoosetheradiotherapyschemeas 1.8 ∼ thestandardfractionationofatotalof54Gyin30fractionsof1.8Gyoveratimerangeof6weeks(5 sessionsperweekfrommondaytofriday). 3. Results 3.1 Computationaldetails Wehavestudiedtheevolutionofthetumordiameterusingourmodelequations(2.2)and(2.3). Tosolve thepartialdifferentialequationswehaveusedstandardsecondorderfinitedifferencesbothintimeand space. Since the tumor diameterin the frameworkof this modeltendsto grow linearlyin any spatial dimensionwe have focusedon the simplest one-dimensionalversionof the model. We havechecked with simulationsin higherdimensionsthatthedynamicsisessentiallythesame andthushavesticked to the simplest possible model. To avoid boundaryeffects and focus on the bulk dynamics, we have assumedourcomputationaldomaintobemuchlargerthanthetumorsize. In each simulation we have computed the tumor diameter as the distance between the points for which the density reaches a critical detection value C that provides a signal in the T2 (or FLAIR) th MRI sequence. Althoughwhichisthatprecisevalueisa debatedquestionandin factdependsonthe thresholdsusedintheimages,wehaveassumedthatC 0.05 0.07. Thisisinagreementwiththe th ∼ − reported value of cellular density about0.16 for detection (Swansonetal., 2008) and a normaltissue densityofabout0.1. Inagreementwithpreviousstudieswetakeafataltumorburden(FTB)sizeof6 cmindiameter(Swansonetal.,2008;Wangetal.,2009). Asparameterscontainingusefulinformation we have computed: the time in which the tumor starts regrowing after the therapy, usually called in 8of26 V.M.PE´REZ-GARC´IAetal. 60 (a) ½ = 0.00712 day-1 50 ) m m 40 ( r te ½ = 0.00356 day-1 e m 30 a i D r o 20 m u T 10 0 0 1 2 3 4 5 Time (years) 0.5 (b) 0.4 A(t) 0.3 B(t) 0.2 0.1 0.0 0 40 80 120 160 200 Time (days) FIG.1.(a)Tumordiameterevolutionfortwodifferentvaluesoftheproliferation:r =0.00356day−1,r =0.00712day−1.Other parametervaluesare: D=0.0075mm2/day,criticaldetectionvalueCth=0.07,SF1.8=0.90andk=1. Theinitialtumorcell densitiesaretakenasCd(x,0)=0,C(x,0)=0.4sech(x/6)withxmeasuredinmm,whatgivesaninitialtumordiameterof28.8 mm. Radiotherapy followsthestandardscheme(6weekswith1.8Gydosesfrommondaytofriday)andstartsattimet=0. Circlesdenotemeasurementseverythreemonthsthatwouldcorrespondtoaclosefollow-upofthepatient.Theupperdashedline (horizontal)showsthefataltumorburdensizetakenthroughthispapertobe6cmasdiscussedinthetext. (b)Evolutionofthe tumorcellamplitudesA(t)=maxx C(x,t) andB(t)=maxx Cd(x,t) duringthefirst200daysshowingtheearlyresponsetothe therapyforr =0.00356day−1. | | | | DELAYEFFECTSINRADIOTHERAPYOFLGG 9of26 10 (a) 8 ) y ( 6 e m Ti Survival time 4 2 Growth delay Time to tumor progression 0 0.002 0.004 0.006 0.008 0.01 ½ (day- 1 ) m) 0 m (b) ( n -4 o ti c u d -8 e r r e t -12 e m a Di -16 x a M -20 0.0 0.002 0.004 0.006 0.008 0.01 ½ (day- 1 ) FIG.2. Dependenceofthe(a)survivaltime(solidline),growthdelay(dashedline),timetotumorprogression(dottedline)and (b)maximalreductionindiameterasafunctionofr . Thecurvessummarizetheoutcomeofmanyindividualsimulationswith initialdataCd(x,0)=0,C(x,0)=0.2exp x4/81920 withxmeasuredinmmthatgivesaninitialtumordiameterof33.80mm. − Radiotherapyfollowsthestandardscheme(6weekswith1.8Gydosesfrommondaytofriday)andstartsattimet=0. Other (cid:0) (cid:1) parametersusedinthesimulationsareasinFig.1. 10of26 V.M.PE´REZ-GARC´IAetal. clinical practice time to tumor progression(TTP), the time for which the tumor size equals its initial size-denotedasgrowthdelay(GD)-andthetimeforwhichthetumorsizeequalstheFTBorsurvival time(ST). Wehavestudiedabroadrangeofparametervaluescorrespondingtothepossiblerangeofrealistic valuesintheframeworkofoursimpledescriptionofthetumordynamicsanditsresponsetoradiother- apy. We have also taken differenttypesof initial data rangingfrom morelocalized (such as gaussian initial profiles) to more infiltrative (such as sech type functions). In what follows we summarize our results. 3.2 Tumorproliferationratedeterminestheresponsetoradiotherapy Inafirstseriesofsimulationswehavestudiedthedependenceoftheevolutionofthetumordiameteron theproliferationrate. Fig. 1(a)showstheevolutionofthetumordiameterfortwodifferentproliferation rates r =0.00356 day 1 and r =0.00712 day 1 (Fig. 1). In the first case of low proliferation, the − − tumorrespondsmoreslowlytotherapyasmeasuredbythespeedoftumorregression(decreaseinsize) butthetotalresponsetimeissignificantlylonger,beingthe timetotumorprogressionof16.9months against8.2monthsinthelatercase. Alsothegrowthdelaysare14.7monthsforthefasterproliferating tumor against29.9monthsfor the less proliferativeone. Finally the “virtualpatient” with the slowly proliferating tumor survives much longer than that with the more proliferative one. This is just an exampleofatendencyshowninallofoursimulationswheremoreaggressivetumorsrespondearlierto thetherapy. It is remarkablethat this fact is in fullagreementwith the results fromPalludetal. (2012b). Our model based on reasonable biological assumptions leads to a long remission time (e.g. in Fig. 1 of about 3 years) much larger than the treatment duration (6 weeks) and negatively correlated with the tumor proliferationrate. As a second relevant finding, also seen in the results shown by Palludetal. (2012b),weobservethattumorsrespondingfasterhavealsoshorterregrowthtime. Fig 1(b) shows the dynamics of the maximum density of tumor cells (A(t)=max C(x,t)) and x damagedtumorcells(B(t)=max C (x,t)). Ascouldbeexpectedtheamplitudeoffunctionallyalive x d tumorcellsdecreasesduringthetherapywiththeexceptionofthebreaksintheweekendswereasmall increaseisseenandcorrespondinglytheamplitudeofdamagedtumorcellsgrowsaftereveryirradiation andforthefulltreatmentperiod(6weeks=42days). Aftert =42daysthepopulationoftumorcells starts a slow recovery while the population of damaged cells declines in a much longer time scale. However, the width of the total tumor population evolves only in the slow time scale and does not displayanyeffectsduringthetreatmentperiod. Itisimportanttoemphasizethatthisbehaviorisnottheresultofafortunatechoiceoftheparameters but a generic behavior as we have confirmedthrougha large numberof simulations coveringthe full clinically relevantparameterspace. As an example, in Fig. 2 we show how the variationof r overa broadrangeofvaluesleadstothesameconclusion. Largerproliferationvaluesacceleratetheresponse butleadtoearlierregrowthandassuch,shortergrowthdelaysandsurvivaltimes[Fig. 2(a)].Oursimu- lationsalsopointoutthatthemaximumvolumereductionisonlyweaklydependentontheproliferation rater [Fig. 2(b)],thesmallertheproliferationratesthelargerthemaximumreductionindiameter.This factisalsoinverygoodagreementwiththeresultsofPalludetal.(2012b)(seee. g. Fig. 2bottomof theirpaper).