(3.3.11)--脑科学与影像新技术.pdf

Preoperative dynamic contrast-enhanced MRI correlates with molecularmarkers of hypoxia and vascularity in specific areas of intratumoralmicroenvironment and is predictive of patient outcomeRandy L.Jensen,Michael L.Mumert,David L.Gillespie,Anita Y.Kinney,Matthias C.Schabel,and Karen L.SalzmanDepartmentofNeurosurgery,ClinicalNeurosciencesCenter(R.L.J.,M.L.M.,D.L.G.),DepartmentofInternalMedicine,DivisionofEpidemiology(A.Y.K.),Department of Radiology,Utah Center for Advanced Imaging Research(UCAIR)(M.C.S.),Department of Radiology,Division ofNeuroradiology,Clinical Neurosciences Center,Universityof Utah,Salt Lake City,Utah(K.L.S.);AdvancedImaging Research,Oregon HealthScience University,Portland,Oregon(M.C.S.);Huntsman Cancer Institute,Salt Lake City,Utah(R.L.J.,D.L.G.,A.Y.K.,K.L.S.)Corresponding author:Randy L.Jensen,MD,PhD,Huntsman Cancer Institute and Departments of Neurosurgery,Radiation Oncology,OncologicalSciences,Clinical Neuroscience Center,University of Utah,175 North Medical Drive,Salt Lake City,Utah 84132(randy.jensenhsc.utah.edu).Background.Measures of tumor vascularity and hypoxia have been correlated with glioma grade and outcome.Dynamic contrast-enhanced(DCE)MRI can noninvasively map tumor blood flow,vascularity,and permeability.In this prospective observational cohortpilot study,preoperative imaging was correlated with molecular markers of hypoxia,vascularity,proliferation,and progression-freeand overall patient survival.Methods.Pharmacokinetic modeling methods were used to generate maps of tumor blood flow,extraction fraction,permeability-surfaceareaproduct,transferconstant,washoutrate,interstitial volume,bloodvolume,capillarytransittime,andcapillaryheterogen-eity from preoperative DCE-MRI data in human glioma patients.Tissue was obtained from areas of peritumoral edema,active tumor,hypoxic penumbra,and necrotic core and evaluated for vascularity,proliferation,and expression of hypoxia-regulated molecules.DCE-MRI parameter values were correlated with hypoxia-regulated protein expression at tissue sample sites.Results.PatientsurvivalcorrelatedwithDCEparametersin2cases:capillaryheterogeneityinactivetumorandinterstitialvolumeinareasof peritumoral edema.Statisticallysignificant correlations were observed between several DCE parameters and tissue markers.In add-ition,MIB-1indexwaspredictiveofoverallsurvival(P.044)andcorrelatedwithvascularendothelialgrowthfactorexpressioninhypoxicpenumbra(r 0.7933,P.0071)and peritumoral edema(r 0.4546).Increased microvessel density correlated with worse patientoutcome(P.026).Conclusions.Our findings suggest that DCE-MRI may facilitate noninvasive preoperative predictions of areas of tumor with increasedhypoxiaandproliferation.Bothimagingandhypoxiabiomarkersarepredictiveofpatientoutcome.Thishasthepotentialtoallowunpre-cedented prognostic decisions and to guide therapies to specific tumor areas.Keywords:DCE-MRI,HIF-1,hypoxia,vascularity,VEGF.Glioblastoma(GBM)is an aggressive primary brain tumor inhumans with an estimated 2-year survival rate of only 0%5%despiteaggressivetreatment.GBMisoneofthemosthighlyvascu-larized of human tumors,but its microcirculation is functionallyvery inefficient compared with that of the normal brain.1,2Studies have demonstrated that the extent of necrosis correlatesinversely with patient outcome and survival.3Intratumoral necro-sis and vascular endothelial proliferation are histological featuresof GBM that are known to distinguish low-grade from high-gradegliomas.4Neitherwhattriggersthistransformationnorthemech-anism by which this is accomplished are known.Tumorhypoxiaisacriticalfactorthatinfluencestumorresponseto radiation therapy and some chemotherapy agents.5The oxy-genation status of a tumor is also a factor in the regulation ofgene expression for malignant progression of tumors.6Malignantbrain tumors are thought to have large proportions of hypoxictissue that contribute to resistance to radiation and chemother-apy.Magnetic resonance imaging(MRI)has emerged as themost powerful tool in the diagnosis of various central nervoussystemdisorders.ConventionalMRIprovidesimportantanatomic-al and diagnostic information for brain tumors.7,8Althoughgadolinium-based anatomic MRI is routinely used to predict theReceived 4 June 2013;accepted 10 August 2013#The Author(s)2013.Published by Oxford University Press on behalf of the Society for Neuro-Oncology.All rights reserved.For permissions,please e-mail:.Neuro-OncologyNeuro-Oncology 16(2),280291,2014doi:10.1093/neuonc/not148Advance Access date 4 December 2013280gradeofagliomaandprovidesimportantanatomicalanddiagnos-tic information,postoperative histological grade or tumor type isoften different from that predicted by imaging alone.9Dynamiccontrast-enhanced(DCE)MRIiscapableofquantitativelymeasur-ing tissue blood flow,vascularity,and parenchymal contrastuptakeviakineticmodelingmethodsandhasbeenusedintheas-sessment of gliomas,especially GBM.1019These techniques havealsobeenusedtoevaluatetheoxygenationstatusoftumors.2022Molecular markers of hypoxia include the transcription factorhypoxia-inducible factor-1a(HIF-1a)as well as other hypoxia-regulated proteins such as vascular endothelial growth factor(VEGF),carbonic anhydrase IX(CA-IX),and glucose transporter-1(GLUT-1).2325Weandothershaveshownthattherearecorrelationsbetween brain tumor grade,vascularity,and HIF-1a expressionbased on a small series of brain tumors.9,2628Although othershave reported hypoxia markers predictive of patient outcome in anumber of tumor types,2931we have been unable to find anyas-sociation of hypoxia biomarkers and patient outcome.9It is pos-sible that there is differential expression of these biomarkers inspecific microenvironments within a given tumor.Thus,randomsampling of a tumor might not reflect expression of thesemarkers in a meaningful manner.We hypothesize that measure-ment of these hypoxia markers in specific,well-defined areas ofthe tumor may be more predictive of patient outcome.Further-more,we hypothesize that preoperative imaging from theseareas might also prove useful for predicting patientoutcome non-invasively.In this pilot study,presurgical imaging studies were directlycorrelated with tissue taken from specific areas of a given tumor.Using an intraoperative navigation system,tumor tissue wastaken from 4 distinct tumor regions for analysis.These areasincluded presumednecroticareas(NC,heterogeneous nonenhan-cing tumor core),presumed hypoxic penumbra(HP,enhancingarea immediately surrounding areas of necrosis),active tumor(AT,nodular enhancing tissue at the outer edge of the tumor),and peritumoral edema(PE,nonenhancing area surroundingthe tumor,which appears bright on T2-weighted imaging)(Fig.1).Each of these areas was evaluated for the expression ofhypoxia-regulated molecules as well as tumor vascularity andtumor proliferation(MIB-1 labeling index).Expression levels ofthe various markers in these samples were correlated withimaging biomarkers derived from DCE-MRI data from spatiallycolocated regions of interest,as well as with tumor behavior andpatient outcome.MethodsPatient Selection and EnrollmentPatients with a newly suspected malignant glioma,which was identifiedby MRI as having a single focus at least 2 cm in cross-sectional diameterand considered to be surgically resectable at the time of presentation,were approached about the study.After a thorough discussion,writtenconsent was obtained from each patient for this University of UtahInstitutional Review Board-approved protocol.Measures of Patient OutcomeCompleteresectionwasindicated byremoval of all gadolinium-enhancingtumortissue(oralltissueshowingT2signalabnormalityfornonenhancingtumors).Progression was defined as the time at which management wasfirst altered from that initially begun for the patient.For most cases,thisFig.1.Schemaoftumorareasassayedinthisstudy.Tumortissuewastakenfrom4areasforspecificimaginganalysis.Theseareasincludedthepresumednecrotic core(heterogeneous nonenhancing tumor core),presumed hypoxic penumbra(enhancing area immediately surrounding areas of necrosis),active tumor(nodular enhancing tissue at the outer edge of tumor),and peritumoral edema(nonenhancing area surrounding tumor,bright onT2-weighted imaging).Jensen et al.:DCE-MRI and hypoxia biomarkers predict outcomeNeuro-Oncology281waswhen therewas eitherunequivocalincrease in fluid-attenuated inver-sionrecovery(FLAIR)/T2signalabnormalityornewlydetectedareasofcon-trastenhancementonfollow-upMRIrequiringfurthersurgery,radiation,orchemotherapy.Progression-free survival(PFS)was calculated from thedate of initial imaging to documented progression.Overall survival(OS)was calculated from the date of initial imaging to documented deathfrom the Social Security Death Index.There was no central review ofimaging.Preoperative Patient ImagingPatients who satisfied all of the eligibility criteria underwent standard pre-operative contrast-enhanced anatomic MRI and DCE-MRI.Up to 4 distinctsites were identified for specific imaging analysis to provide resectiontargets:(i)presumed NC,(ii)presumed HP,(iii)AT,and(iv)PE.In sometumors,suchashomogeneouslyenhancinglesionsorminimallyenhancinglesions,itwasnotpossibletoidentifyall4regions.Allimageswereobtainedsuchthattheycouldbeloadedintotheintraoperativenavigationsystem,asper protocol for any patient undergoing surgical resection(Fig.2).DCE-MRI Data AcquisitionT1-weighted DCE-MRI data were acquired as described previously.32Afterstandard noncontrast imaging for preoperative planning was completed,the lesion of interest was identified(Fig.2),and maps of precontrast longi-tudinal relaxation time(T10)were measured using the variable flip anglespoiled gradient echo(SPGR)method with flip angles of 38 and 208.33DCE-MRI measurements were then performed using a fast 3-dimensionalSPGR sequence on a 1.5T Siemens TIM Avanto scanner or a 3T SiemensTIMTrioandVerioscannerswiththesamefieldofviewusedforthevariableflip angle measurements.Pulse sequence parameters were chosen tomaximize sampling rate within constraints of adequate signal-to-noiseratio and coverage of the lesion of interest.Because DCE-MRI data wereacquired as add-on measurements to clinically indicated stereotacticimaging prior to brain surgery,imaging protocols varied slightly to accom-modate different scanner hardware and specific absorption rate limits.Astandard(0.1 mmol/kg)doseoflow-molecular-weightgadoliniumchelate contrast agent(Omniscan,GE Healthcare or Multihance,BraccoDiagnostics)was injected over 4 seconds through an 1822 gauge intra-venous catheter into the antecubital vein,followed by a 20 mL salineflush,injected at 2 mL/s using a power injector(Medrad Spectris Solaris).Contrast injection was timed to coincide with the end of the 10th frameof dynamic data.Data were acquired at an isotropic spatial resolution of2 mm for a total imaging time of at least 6 minutes with temporal reso-lution of 5 seconds per frame or better.A 12-channel transmit/receivehead coil was used.Flip angle variation along the slab-encoding directionwas estimated using a homogeneous phantom of known T10,and,in 3study patients,by assuming a constant T10value of 240 milliseconds forsubcutaneous fat.These experiments indicated that the impact of flipangle variation near the slab boundaries was small,except for roughlythe outer 10%of slices at each boundary.To minimize the potential forbias in conversion of signal to contrast concentration,34these slices wereexcluded from our analysis.Precontrast signal intensity,S0(t),was deter-mined by averaging the 10 baseline images,and signal-to-noise ratiowas computed from the ratio of the baseline signal to its standarddeviation.Time curves of relative enhancement were generated from thetime-dependent DCE-MRI signal,S(t),as S(t)=S(t)S0(t)/S0(t).Tissueconcentration-time curves,Ct(t),were then computed by numerical solu-tion of the full nonlinear concentration dependence of the SPGR signal inthe fast exchangelimit,withvoxelwise tissueT10values and flip angle cor-rected forslabprofileeffects.34Literaturevalues forther1and r2relaxivity,appropriate to the administered contrast agent at the imaging fieldstrength,were used.35Concentration measurement uncertainties werecomputed as described by Schabel and Parker.34The direct effect of T20on concentration measurements was eliminated by the use of relative en-hancement;the very short echo times used in imaging should result inminimal susceptibility-induced signal loss.A T20value of 50 millisecondswas assumed in calculations of contrast concentration uncertainty.DCE-MRI Pharmacokinetic ModelingMeasured concentration-time curves,representing tissue uptake of con-trastagentas determinedfrom the DCE-MRIdata,were fit using nonlinearregression modeling to the Gamma Capillary Transit Time(GCTT)model,which is a distributed parameter model that incorporates both contrasttransit through the tumor microvasculature and uptake by the tumor par-enchyma.36In addition to our prior studies in gliomas,this model hasshown utility in analysis of tumors in a rat model.37Applying the GCTTmodel to our data provided spatial maps of imaging biomarkers includingtumor blood flow(F),extraction fraction(E),permeabilitysurface areaproduct(PS),transfer constant(Ktrans),washout rate(kep),interstitialvolume(Ve),bloodvolume(Vb),capillarytransittime(tc),andcapillaryhet-erogeneity(a1).Mean parameter values were calculated by averagingthese maps over spherical regions of interest 5 mm in diameter(roughlycorresponding to the volume of sampled tissue)that were manually core-gistered with sampling sites reported by the MRI-guided intraoperativenavigation system.Tumor Resection,Tissue Procurement,and PatientTreatmentSurgerywasperformedwithin24hoursoftheimagingstudies.Stereotacticpost-contrast MRI images were loaded on the intraoperative navigationsystem and used to select the desired regions of tumor(Fig.2).Tumorwasresectedbytheseniorauthorwiththeintentionofgrosstotalresectionwhenpossible.Afrozensectionexaminationofthetumorwasperformedtoconfirm histological diagnosis.The tumor was removed using the intrao-perative navigation system,which allows for documentation and correl-ation of the location of the tissue removed with the preoperative imagingstudies.All efforts were made to sample the tumor areas of interestbefore any“brain shift”could occur from cerebrospinal fluid removal orbrain relaxation.Only tumors 2 cm in diameter or larger were included inthis study to ensure that the small amount of tumor used as part of thisstudy would not interfere with the ability of the neuropathologist to makean accurate diagnosis.The majority of the tumor was sent for standardpathological examination and tumor grade determination according tothe WHO classification system.4Remaining tissue was immediately snapfrozen and stored in liquid nitrogen until time of analysis.Allpatientsweretreatedwithradiationandchemotherapybasedontheirspecific diagnosis under the guidance of neuro-oncologists.Patients under-went routine blood