Quantitative Assessment of MRI T2 Response to Kainic

Acid Neurotoxicity in Rats in vivoSerguei Liachenko1 Jaivijay Ramu Tetyana Konak Merle G Paule andJoseph Hanigdagger

Division of Neurotoxicology National Center for Toxicological Research US Food and Drug AdministrationJefferson Arkansas and daggerOffice of Testing and Research Center for Drug Evaluation and Research US Foodand Drug Administration White Oak Maryland1To whom correspondence should be addressed at National Center for Toxicological Research US Food and Drug Administration 3900 NCTR RdJefferson AR 72079 Fax (870) - 543 - 7745 E-mail Sergueiliachenkofdahhsgov

ABSTRACT

The aim of this study was to assess quantitative changes in T2 relaxation using magnetic resonance imaging approaches inrats exposed to kainic acid to assess the utility of such endpoints as biomarkers of neurotoxicity Quantitative T2 mappingwas performed in 21 rats before and 2 24 and 48 h after a single ip injection of 10 mgkg of kainic acid Three methods ofquantifying T2 changes were explored (1) Thresholding all voxels exhibiting T272 ms were designated normal tissuewhereas voxels exhibiting T2gt72 ms were designated as lesioned tissue (2) Statistical mapping T2 maps obtained aftertreatment were statistically compared with averaged ldquobaselinerdquo maps voxel-by-voxel (3) Within-subject difference frombaseline for each individual the baseline T2 map was subtracted from the T2 map obtained after treatment Based on thefollow-up histopathological response there were 9 responders 7 nonresponders and 5 animals were not classified due toearly sacrifice at 2 h which was too soon after treatment to detect any morphological evidence The ldquothresholdingrdquo method(1) detected differences between groups only at the later time point of 48 h the ldquostatistical mappingrdquo approach (2) detecteddifferences 24 and 48 h after treatment and the ldquowithin-subject difference from baselinerdquo method (3) detected statisticallysignificant differences between groups at each time point (2 24 and 48 h) T2 mapping provides an easily quantifiablebiomarker and the quantification method employing the use of the same animal as its own control provides the mostsensitive metrics

Key words MRI rat brain registration toxicity kainic acid quantification T2 map

Pharmaceutical and other chemical companies spend a signifi-cant amount of their resources toward evaluating and demon-strating the safety of their new compounds The considerablepreclinical and clinical data generated during the drug develop-ment process undergo rigorous and comprehensive reviews byregulatory agencies such as the Food and Drug Administration(FDA) before a drug is deemed safe for human use (Gad 2014)Great savings could be realized if the drug safety issues could beidentified in the early preclinical stages of drug developmentMany preclinical safety assessment routines rely solely onvisual examination of 2D histopathological slices of tissueand organs from animals treated with the compound under

development In assessing potential neurotoxicity it is ex-tremely difficult to investigate a complex organ such as thebrain using 2D slices alone and the number of tissue slices ex-amined is typically grossly insufficient to cover the entirety ofthe brain This results in the very real possibility that smallareas of damage can easily be missed (Switzer et al 2011) anddemonstrates the significant limitations associated with histo-pathological assessments alone In this context the availabilityof a noninvasive central nervous system (CNS)-specific bio-marker should greatly facilitate the maintenance of a sustain-able pipeline of new drug candidates Employing suchtechnologies will enhance the safety and also lower the cost

Published by Oxford University Press on behalf of the Society of Toxicology 2015 This work is written by US Government employees and is in the publicdomain in the US

1

TOXICOLOGICAL SCIENCES 2015 1ndash9

doi 101093toxscikfv083Advance Access Publication Date April 21 2015

ToxSci Advance Access published June 5 2015 at FD

A L

ibrary on June 12 2015httptoxscioxfordjournalsorg

Dow

nloaded from

associated with new drug development Neuroimaging in par-ticular magnetic resonance imaging (MRI) holds great promisein this regard (Borsook et al 2011 Peterson et al 2011)Recently we have demonstrated the utility of a noninvasive 3Dquantitative MRI T2 mapping approach to guide conventionalpreclinical neuropathological investigations (Hanig et al 2014)T2 mapping is a relatively simple approach available not only tothe high magnetic field research community but also to clinicalsettings (Deoni et al 2003 Filippi et al 2014) There have beennumerous attempts to use T2 relaxation imaging to analyze theneurotoxic effects of various chemical substances (Coley et al1999 Eichler et al 2007 Mavroudis et al 2006 Preece et al 2004Sandoval et al 2003) including those associated with the exci-totoxic agent kainic acid (KA) (Acosta et al 2012 Ebisu et al1996 Hsu et al 2007 Nakasu et al 1995b Pirttila et al 2001Righini et al 1994 Wolf et al 2002) These studies reportedmixed results ranging from no changes (Acosta et al 2012 Ebisuet al 1994) to significant increases in T2 signals (Hsu et al 2007Nakasu et al 1995b Righini et al 1994) with positive correlationsto histopathology (Pirttila et al 2001) However most of thesestudies did not utilize quantitative T2 maps but instead dealtwith relative changes in image intensity of T2-weighted imageswhich may have affected the sensitivity of the approach

In this article we present a comparison of severalapproaches to the quantification of parametric T2 maps in a ratmodel of acute neurotoxicity caused by KA and identify anoptimal quantification method for further development of MRIbiomarkers of neurotoxicity

MATERIALS AND METHODS

AnimalsAnimal use protocol was approved by the National Center forToxicological Research (NCTR) Institutional Animal Care andUse Committee Thirty-seven male Sprague-Dawley rats(Charles River Labs 364 6 40 g 67 6 4 days old) were used in thisstudy Animals were kept under a typical 1212 h daynight lightcycle fed ad libitum on the standard rat chow diet and single-housed in polypropylene cages outfitted with ventilated-topisolators and natural wood chip bedding Twenty-one ratsreceived a single dose of KA (10 mgkg ip) and 16 animalswere treated with a single dose of the vehicle alone (saline2 mlkg ip)

Magnetic Resonance ImagingMRI and spectroscopy were performed using a 7 T30 BiospecAV III MRI system (Bruker MRI Billerica Massachusetts) with a12 cm internal diameter (ID) gradient insert (600 mTm) A 4channel array rat brain radiofrequency (RF) coil (Bruker MRI)was used for receiving and a 72 mm birdcage RF coil (BrukerMRI) was used for transmitting the MRI signals Animals wereanesthetized with isoflurane (3 in an induction chamber fol-lowed by 1ndash2 via a nose cone for maintenance at 1 lmin flowrate in oxygen) Body temperature was kept at 373C 6 06Cusing a water-heated imaging cradle Animals were fittedwith an IP catheter for the delivery of KA or saline whereasinside the MRI after the completion of baseline scan Imagingwas repeated at 2 h and some animals were reimaged at 24(Nfrac14 10) or 48 h (Nfrac14 21) after KA delivery For T2 relaxationmapping of the whole brain a multi-echo spin echo sequencewas used (matrix 192 192 number of slicesfrac14 24 fieldof viewfrac14 384 384 cm slice thicknessfrac14 1 mm echospacingfrac14 15 ms 16 echoes relaxation timefrac14 6 s number ofaveragesfrac14 1 total acquisition time approximately 10 min)

Image Process and AnalysisT2 maps were calculated from a set of 16 echo images usingvoxel-by-voxel fitting of image intensities to the single expo-nential function

M frac14 M0 expethTE=T2THORN

where M is image intensity in a given voxel (dependent varia-ble) M0 is the maximal image intensity of the same voxel at thezero echo time (proton density image) TE is the echo time ofthe particular echo image (independent variable) and T2 is thetransverse relaxation time (sought parameter) The resultant T2

maps were arbitrarily but consistently color coded to conveytheir quantitative nature as shown in Figure 1 Nonbrain tissuewas removed from the T2 map images using a semiautomatedskull stripping procedure Images were then co-registered to thesame template image using a rigid body surface registrationprocedure without interpolation using Analyze 110 software(AnalyzeDirect Inc Overland Park Mississippi) The templatefrom a single rat having subjectively the best image quality andorientation was used as the co-registration template T2 mapswere smoothed using a 2D running averages filter as followsthe value of each in-plane pixel of the T2 map was set equal toits average value and the 8 surrounding pixels Maps were thendownsampled to half of the original resolution in each in-planedimension using linear interpolation and keeping the numberof slices intact Three methods of quantification of KA-inducedT2 changes were employed

Thresholding All T2 maps were stripped of all tissue except thebrain (skull-stripped) and analyzed using an arbitrary thresholdvoxels with T2 values 72 ms inclusive were designated as nor-mal tissue and voxels with T2 values above 72 ms were desig-nated as lesioned tissue

Statistical mapping All T2 maps were skull-stripped co-registered 2D-smoothed and downsampled in that particularorder T2 maps of the rat brain before treatment (baseline)were used to produce average and standard deviation maps T2

maps of the rat brain after treatment with KA or salinewere then compared voxel-by-voxel using Studentrsquos t test toidentify those voxels with statistically significant increases inT2 that were produced by the treatment To decrease thenoise effects only voxels in which the coefficient of variance(CV) values for the averaged baselines werelt 10 were testedA 1-sided t test with Bonferroni correction was employed toaccount for multiple comparisons We have used a 1-sided ttest here because even though the exact mechanism ofapparent tissue T2 changes during neurotoxic events is poorlyunderstood the available literature suggests only increased T2

values in lesioned brain tissue (Ebisu et al 1996 Hsu et al 2007Lee et al 2010 Mavroudis et al 2006 Nakasu et al 1995bPreece et al 2004 Righini et al 1994) This observation could beat least partially related to edema andor loss of cellularintegrity

Within-subject difference from baseline All T2 maps were skull-stripped coregistered 2D-smoothed and downsampled as inprevious methods The coregistered baseline T2 map was sub-tracted from the T2 maps from the same animal obtained at dif-ferent time points after treatment To decrease the noise effectsonly voxels with CV values in the averaged baseline map (thismap was the same one used in the previous methods) oflt 10and with changes in T2 values ofgt 10 ms were counted

2 | TOXICOLOGICAL SCIENCES 2015

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

Negative changes were assumed to be due to noise and notincluded in the analyses

The sensitivities of the markers computed using theaforementioned analytical methods were estimated as probabil-ities of positive MRI signals given that lesions were present andthe specificities were estimated as the probabilities of negativeMRI signals given the absence of lesions (Altman and Bland1994) The MRI signal was considered positive (or negative) ifthe value derived from the respective analysis for a given indi-vidual subject was outside (inside) the confidence interval for acorresponding control (between-subject measure) or baseline(within-subject measure) at an alpha of 05 The presence of alesion was determined by histopathologymdasha lesion was consid-ered present if any region of a given subjectsrsquo brain was charac-terized by a pathology score higher than normal

HistologyAt the end of the experiment (2 24 or 48 h after KA or salineadministration) animals were deeply anesthetized with 5 ofisoflurane and perfused transcardially with 4 paraformalde-hyde in cacodylate buffer for histopathological analysis accord-ing to specifications provided by Neuroscience Associates Inc(Switzer 2000) Each brain was dissected into 80 sections (40 mmthick) which were stained with amino CuAg and neutral redbackground (DeOlmos and Ingram 1971 Switzer 2000) Damageassessment of the brain tissue was scored utilizing a relativeneurodegeneration system (0mdashno degeneration 1mdashminimal2mdashmild 3mdashmoderate 4mdashmarked degeneration)

StatisticsStatistical analysis of the differences across time and acrossexperimental groups was performed using a 2-factor ANOVAand a 1-factor repeated measures ANOVA design depending onthe structure of the data with post hoc Holm-Sidak multiplegroup comparisons Effect size was calculated using Cohenrsquos dapproach (Cohen 1988) Data are shown as means 6 SEM

RESULTS

Treated animals showed varying degrees of response to KAadministration as judged by both MRI and histology rangingfrom no response to severe changes (Fig 1) Thereafter only res-ponder animals were chosen for further analysis A rat was con-sidered a responder if histological assessment of any part of thebrain resulted in a pathological score gt0 Out of 21 treated rats9 animals were classified as positive responders 7 as nonres-ponders and 5 were not classified because they were sacrificed2 h after the treatment which was too early for the expressionof morphological evidence (Switzer 1991 2000 Switzer et al2011)

MRI AnalysisThresholding Figure 2 shows the combined T2 histogram plot fordata obtained from a single slice of a representative rat brainbefore KA administration (gray area) and similar plot of a man-ually segmented out region (amygdala and surrounding area)showing marked T2 enhancement in the same rat 48 h after KAadministration (black area) Based on these histogram plots thethreshold of 72 ms was empirically selected as the boundarybetween normal tissue and tissue lesioned by KA injection Thenumber of voxels below (designated as ldquonormalrdquo) and above(designated as ldquolesionedrdquo) this threshold were counted and con-verted into volumes for all responders Voxels having inten-sities higher than 122 ms were presumed cerebrospinal fluidand not counted Results are plotted in Figure 3 where it can beseen that the volume of the ldquolesionedrdquo tissue increased in bothcontrol and KA-treated groups 24 h after treatment anddecreased in the control group 48 h after treatment Significantdifferences were observed between control and KA groups atthe 48 h time point There was no significant difference(Pfrac14 057) between responders and non-responders and betweencontrol and nonresponders (Pfrac14 096 not corrected for multiplecomparisons) 48 h after KA administration (data not shown)Due to an insufficient number of non-responders we did not

FIG 1 Examples of T2 maps of an unexposed rat brains (top row) and the brains of the same rats 48 h after KA exposure (bottom row) The severity of changes ranges

from none (A) to medium (B) to severe (C) indicating different responses of the animals to kainic acid Quantitative T2 maps are arbitrarily color coded according to the

color scale on the right Only a single slice from the 24 slices that were captured for each animal to encompass the whole brain is presented here

LIACHENKO ET AL | 3

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

compare their data with those of responders at other timepoints The total volume of the tissue was stable in all groups(2200 6 12 mm3) except for KA-treated animals at 48 h when itincreased significantly (2341 6 45 mm3 Plt 05)

Statistical mapping The results of the voxel-by-voxel statisticalcomparison mapping are depicted in Figure 4 Statistically sig-nificant differences in the volumes of ldquolesionedrdquo tissue weredetected in the KA group at every time point and in the controlgroup at the 24-h time point compared with the baseline

However significant differences between control and KA groupswere detected only at 24 and 48 h after treatmentNonresponders in the KA group showed significantly (Pfrac14 023)smaller ldquolesionrdquo volumes in comparison to responders and nodifference (Pfrac14 071 not corrected for multiple comparisons) incomparison to controls 48 h after treatment (data not shown)

Within-subject difference from baseline The volumes of ldquolesionedrdquotissue detected by this method are shown for all groups inFigure 5 Statistically significant differences between controland KA groups were detected at all observed time points Also

FIG 2 Histograms showing the distribution of T2 values for single slices of brain images before KA treatment (A and gray area on graph) and a manually segmented

region [B (showing an areamdashoutlined in graymdashencompassing the amygdala) black area on graph] from the same slice of the brain of the same animal 48 h after KA

administration Based on this plot the arbitrary T2 threshold for distinguishing ldquonormalrdquo and ldquolesionedrdquo tissues was set at 72 ms

FIG 3 The change in volume of the ldquolesionedrdquo tissue calculated using the

thresholding method (see Materials and Methods section) in responder rats

before (0 h) and after (2 24 and 48 h) saline (control gray bars) or KA (black bars)

administration Data for the KA group at 24 and 48 h are for responders only

Data are means 6 SEM significantly different from time 0 within the same

group (Plt 05) significantly different between groups at the same time point

(Plt 05)

FIG 4 The change in volume of the ldquolesionedrdquo tissue calculated using the statis-

tical mapping method (see Materials and Methods section) in rats before (0 h)

and after (2 24 and 48 h) saline (control gray bars) or KA(black bars) administra-

tion Data for the KA group at 24 and 48 h are for responders only Data are

means 6 SEM significantly different from time 0 within the same group

(Plt 05) significantly different between groups at the same time point (Plt 05)

4 | TOXICOLOGICAL SCIENCES 2015

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

nonresponders in the KA group showed significantly (Pfrac14 018)smaller ldquolesionrdquo volume in comparison to responders and nodifference (Pfrac14 613 not corrected for multiple comparisons) incomparison to controls 48 h after treatment (data not shown)

Comparison of the relative performance of T2 MRI analyticalapproaches Quantitative comparisons of the effectiveness andperformance of the 3 employed image analysis methods usingCohenrsquos d effect size metrics (Cohen 1988) are presented inTable 1 Estimates of the sensitivity and specificity of thesemethods based on the limited data set are also listed in thetable The within-subject difference from baseline methodshows the largest effect for both between- and within-subjectmeasurements (all numbers weregt 08) and the smallest effectsizes for the thresholding method small (dlt 05) for between-subject measurements (2 and 24 h) and small to medium(05lt dlt 08) for within-subject measurements (48 and 2 hrespectively) The effect size for the statistical mapping methodwas small only at 2 h for between-subject measurements

HistopathologyFigure 6 shows examples of the neuropathological changesobserved in rats exposed to KA at different times after exposureThe figures in the top row are anatomical reference MRI scanswith superimposed areas showing the T2 value changes (T2 acti-vation maps) as detected using the within-subject differencefrom baseline method Figures in the second row of the figureare histological samples stained with amino CuAg (DeOlmosand Ingram 1971 Switzer 2000) from the same rats from whichthe T2 activation maps were generated in the top row Twohours after KA administration a slight increase in the T2 signalcould be seen in the hippocampal area when no silver depositswere detectable in the histology slide However the backgroundstaining (neutral red) revealed disarrayed cellular layers in theCA3 region of the hippocampus which is consistent with edem-atous changes The pathology was graded as 0 (no degeneration)due to the absence of silver stain Twenty-four hours after KAadministration increases in T2 are seen over larger areas andthe corresponding histological damage is evidenced by strongersilver deposition (grade 3) in the CA3 region Forty-eight hoursafter KA insult T2 changes are even more wide spread and the

CA3 region again shows a high level of silver staining (grade 2)Animals which did not show any pathological changes at 24and 48 h after KA administration (nonresponders) also did notshow any statistically significant differences from controls in T2

values as analyzed using all 3 methods described herein

DISCUSSION

Current strategies for assessing the neurotoxicity of new drugsor chemical agents employ multidisciplinary approachesincluding pathological biological physiological and behavioralevaluations (Costa 1996 1998 Slikker 1991 Slikker and Bowyer2005 Slikker et al 1996 Slikker and Gaylor 1995) This is a com-plex approach and thus it might be helpful to develop new toolssuch as biomarkers (Costa 1996 Slikker 1991 Slikker andBowyer 2005) that can serve as surrogate endpoints and speedup toxicity testing during drug discovery and developmentdecrease the number of animals needed for preclinical testingand increase the power of detection of the risks associated withexposure to the new chemical The aim of this study was to fur-ther describe the development of noninvasive MRI biomarkersof neurotoxicity which was only recently proposed (Hanig et al2014) A commonly accepted definition of a biomarker statedthat it is an objectively measured indicator of some normal orpathogenic process (Biomarkers Definitions Working Group2001) The objectivity of such measurement implies the abilityto quantify the biomarker in an operator-independent mannerand a first step toward approaching the formal qualification ofsuch a biomarker by the FDA (Goodsaid and Frueh 2007Woodcock et al 2011) is the detailed assessment of its quantita-tive properties Here using kainic acid-treated rats as an estab-lished model of neurotoxicity (Olney 1990) 3 differentapproaches to the quantification of changes in T2 relaxation inthe rat brain during the development of neurotoxic lesions wereexplored The noninvasive nature of MRI permits repetitivescanning of the same subject over the course of the disease orin our case toxicity progression which allows for observationof individual variations in anatomy physiology and reactivityto the intervention (KA in the present example) Indeed it wasshown that the quantification approach which employs awithin-subject difference from baseline method generated thelargest effect size of the methods used

The simplest and easiest method of the 3 methods chosen isthe thresholding approach which did not require elaborateimage manipulation such as coregistration and filtering It didrequire skull stripping that can be performed automaticallywith minimal manual postprocessing However this methodwas not sensitive to individual variations in T2 among rats andin particular to the size of the ventricular system which con-tains CSF and therefore is a source of the very high T2 valuevariations even in baseline images This method is also lesssensitive to the changes in T2 values over time as it only helpsto segment particular images into areas with high and low T2

intensity without reference to baseline valuesThe second method usedmdashstatistical parametric mappingmdash

involves more elaborate image manipulations in addition toskull stripping These include coregistration smoothing down-sampling and averaging Downsampling and simple smoothingwere needed to counteract noise in the T2 maps which arisesfrom many sources including normal subject physiologicalchanges (respiration blood flow temperature fluctuations)subject movement instrument noise and the error associatedwith voxel-by-voxel fitting of the T2 relaxation Similar techni-ques are widely used in functional MRI experiments to deal

FIG 5 The change in volume of the ldquolesionedrdquo tissue calculated using the

within-subject difference from baseline method (see Materials and Methods sec-

tion) in rats after saline (control gray bars) or KA (black bars) administration

Data for the KA group at 24 and 48 h are for responders only Data are

means 6 SEM significantly different between groups at the same time point

(Plt 05)

LIACHENKO ET AL | 5

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

with these same issues (Monti 2011 Ramsey et al 2002) Thestatistical mapping method was more sensitive to longitudinalchanges in T2 because it was able to identify significant differ-ences between control and KA-treated rats as early as 24 hafter treatment This approach similar to standard functional

MRI (fMRI) techniques also suffers from the large number ofvoxel-by-voxel statistical tests that are needed and hence thesensitivity is decreased due to the correction for multiple com-parisons Individual variations in brain anatomy and especiallyin the anatomy of the ventricles and the brain boundariesmdashthe

FIG 6 Visual correspondence of the changes in T2 values (top row) and histopathology (middle row) of individual rats 2 (A D) 24 (B E) and 48 h (C F) after administra-

tion of kainic acid For A B and C the red mask overlaying the grayscale background template brain image denotes the areas with significant T2 changes as detected

using the within-subject difference from baseline method (see Materials and Methods section) For D E and F amino CuAg stained histological images from the same

animals as earlier A B and C respectively G is a magnification of the image in D and H represents the CA3 layer of the hippocampus (approximately the same area

shown in G) of a control rat Note that although no silver deposits are visible 2 h after KA treatment (D G) the cellular integrity in the CA3 region of hippocampus (G) is

showing some disruption compared with the control animal (H) Scale bars are 300m in D E F and 200m in G and H

TABLE 1 Comparison of the Relative Performance of T2 MRI Quantification Methods

Time h Effect Size (Cohenrsquos d) Sensitivity Specificity

Between-Subject Within-Subject

Thresholding 2 012 056 mdash mdash24 038 093 10 02548 145 039 08 067

Statistical Mapping 2 005 086 mdash mdash24 172 272 10 1048 174 180 10 05

Within-Subject Difference With Baseline 2 111 164 mdash mdash24 214 203 10 0048 148 131 10 06

Effect size was calculated using Cohenrsquos d metrics (Cohen 1988) comparing responders to the control at the same time point (between-subject) and to the baseline

(within-subject) The sensitivity was estimated as probability of positive MRI signal given that lesion was present and the specificity was estimated as the probability

of negative MRI signal given the absence of lesion (Altman and Bland 1994) The MRI signal was considered positive (or negative) if the value derived from the respec-

tive analysis for a given individual subject was outside (inside) the confidence interval for a corresponding control (between-subject measure) or baseline (within-sub-

ject measure) at an alpha of 05 The presence of a lesion was determined by histopathologymdasha lesion was considered present if any region of a given subjectsrsquo brain

was characterized by a pathology score higher than normal The sensitivity and specificity at 2-h time point was not estimated as it is too early to detect the presence

of lesion in histopathology slides

6 | TOXICOLOGICAL SCIENCES 2015

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

highest sources of T2 variationmdashpartially accounted by theincreased standard deviation in the corresponding averagedbaseline voxels Unfortunately this effectively excludes thesevoxels from the analysis and therefore increases the chance offalse negative findings if the damage occurs close to the boun-daries of the brain with the skull or ventricles This problem ismagnified in the case of noticeable deviations in baseline anat-omy such as that seen in Figure 1C

The third approach usedmdashthe within-subject differencefrom baseline methodmdashgives the best results in terms of itsability to detect small T2 changes at very early time pointsmdash2 hThis method also required the spectrum of preprocessing stepsincluding skull stripping coregistration smoothing and down-sampling Although in this study all images were coregisteredto the same coordinates it is only possible to coregister eachbrain image to the baseline image within the same subject Thiscan potentially improve the final coregistration and thus thesensitivity of the measurement of T2 changes especially in thecase of individual deviations in brain anatomy

A comparison of the effect sizes (Table 1) for the 3 analyticalmethods reveals that the third approachmdashthe within-subjectdifference from baseline approachmdashdetects large effects at alltime points (Cohen [1998] considers an effect to be small if itislt 05 medium if it islt 08 and large if it isgt 08 as computedusing his d metrics) The second method (statistical mapping)showed all effect sizes to be large except for the between-sub-ject value at 2 h and the first method showed the smallesteffect sizes The estimates of sensitivity and specificity in ourexperiment are very limited and may not reflect the real per-formance of MRI This is due to the small number of observa-tions and crude estimates of lesion presence in thehistopathology data Much more precise estimates of sensitiv-ityspecificity for MRI would be possible if one could correlatedirectly the MRI findings to histology images to detect the ratesof truefalse negativespositives area-by-area or even voxel-byvoxel However this would require a thorough quantification ofthe histology images to derive quantitative maps of neuronaldegeneration These data are not presently available

We have observed statistically significant increase in theldquolesionedrdquo volumes in the control group at 24 h after KA treat-ment using thresholding and statistical mapping methods (Figs3 and 4) The exact reason for this increase is unknown Wespeculate that repeated anesthesia within 24 h time span mayhave caused such changes which warrants further investiga-tion Note that within-subject difference from the baselinemethod did not show such changes because we subtractedbaseline images from the actual images and therefore no statis-tical comparisons to the baseline were made

In our ldquostatistical mappingrdquo and ldquowithin-subject differencefrom baselinerdquo methods we used an arbitrary statistical thresh-old Namely only those voxels having CV values in the corre-sponding averaged control baseline images oflt 10 wereutilized This threshold was just above the maximum level ofthe CV values obtained for brain areas which were far awayfrom the ventricles and the skull (striatum) This procedureeliminated approximately 45 of voxels characterized by highnatural variability which were those predominantly locatedalong the outside borders of the brain and inside the brainaround the ventricles The high variability seen in these areaswas produced by imperfect registration and a partial volumingeffect due to slight differences in initial positioning of the ani-mals inside the MRI system

Surface registration procedures available in Analyze 110were used here as it was previously shown that they perform

better than voxel registration procedures in the case of rat brainimaging [unpublished data] Also in this particular casebecause an aim was to try to coregister images having signifi-cant variations in intensity in some places (due to KA neurotox-icity) voxel registration procedures are likely to produce largerregistration errors in comparison to surface registrationprocedures

KA produces an inconsistent neurotoxic response rangingfrom virtually no damage to severe neurodegeneration Thegoal of this study was not only to characterize KA-induced neu-rotoxicity but also to quantify the T2 changes that occur duringthe development of neurotoxicity For this purpose we havepresented data only for subjects that showed a positive neuro-toxic reaction To do this reliably the only approach availablewas to assess the neuropathological changes using standardhistology techniques after obtaining all of the in vivo MRI scansOnly 9 rats (43) were identified as responders to KA treatmentas judged by qualitative histopathology assessments Thissomewhat low rate of lesion development may be due to thefact that the rats were treated with KA during MRI scanswhereas under isoflurane anesthesia Isoflurane similarly tosome other anesthetics may have decreased the incidence ofthe KA-induced seizures (Hunfeld et al 2013) which are neces-sary for the development of a detectable lesions (Acosta et al2012) Five animals (24) were not classified with respect totheir response to KA because they were sacrificed 2 h after KAadministration and this time was not enough for the develop-ment of any morphological evidence of toxicity (Switzer 20001991 Switzer et al 2011) However T2 changes were measura-ble 2 h after KA administration (Fig 5) and in some cases werevisible on MRI scans (Hanig et al 2014) In a separate analysis(data not presented) changes in T2 in the non-responder groupwere not different from the changes seen in the control group48 h after treatment Such variability in response to KA treat-ment has been reported in other studies as well (Immonen et al2008)

The significant increase in T2 values observed after KA treat-ment in the responder group peaked at 24 h and was still veryhigh at the 48-h time point which corresponds to the windowof opportunity for conventional histopathology to detect neuro-nal damage in the brain tissue in this neurotoxicity model(Benkovic et al 2004 Hopkins et al 2000) It has been pointedout (Switzer et al 2011) that although damage to the brain tis-sue may be permanent in such events the ability to detect theevidence of neurodegeneration depends greatly upon the tim-ing of the tissue sampling for classical histopathological evalua-tion (Switzer et al 2011) The time window during which suchchanges are detectable is usually measured in several days Inthe case of KA neuronal degeneration is typically not observedin histological assessment later than 5 days after treatment(Acosta et al 2012 Hsu et al 2007) In this regard a noninvasivemethod for evaluating neurotoxic changesmdashlike the one pre-sented in this article and elsewhere (Hanig et al 2014)mdashmaybecome the tool of choice for probing the life cycle of neurotox-icity for emerging new drugs and other chemical substances

KA-induced neurotoxicity has been previously studied usingvarious MRI applications like diffusion-weighted imagingdiffusion tensor imaging manganese-enhanced MRI fMRImagnetic resonance spectroscopy (MRI) etc (Airaksinen et al2010 Immonen et al 2008 Kuchar et al 2006 Laitinen et al2010 Najm et al 1998 Pirttila et al2001 van Vliet et al 2014Wang et al 1996 Wolf et al 2002) A variety of changes wereobserved in the brains of KA-treated animals includingdecreased size of the hippocampus (Wolf et al 2002) decreased

LIACHENKO ET AL | 7

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

water diffusion (Hsu et al 2007 Kuchar et al2006 Nakasuet al1995b Righini et al 1994 Wang et al1996) altered neuro-chemical profiles (Ebisu et al 1994 1996 Najm et al1998 Tanget al 1998) elevated sodium content (Wang et al1996) blood-brain barrier damage (van Vliet et al 2014) increase of brainactivity (Airaksinen et al 2010 2012) and increased neuronalplasticity (Immonen et al2008 Laitinen et al 2010) Severalinvestigations involving T2 changes in KA models of epilepsy(Acosta et al 2012 Ebisu et al 1996 Hsu et al 2007 Nakasuet al 1995b Pirttilaet al 2001 Righini et al 1994) reportedmixed results ranging from no T2 change (Acosta et al 2012) orlittle evidence of T2 change (Ebisu et al 1996) to moderateincreases in T2 (Nakasu et al 1995a Righini et al 1994) In somethere was a degree of correlation of T2 with pathology (Pirttilaet al 2001) However all of these studies estimated relativechanges in signal intensity using T2-weighted imaging insteadof systematic use of the quantitative T2 mapping Also none ofthese studies involved advanced image processing approacheslike coregistration and statistical mapping or baseline subtrac-tion As is evident from the thresholding method (Fig 3) thedynamic range of changes in the T2 signal (or even worse T2-weighted image intensity) is greatly diminished if the baselinevalues are not subtracted or taken into account by some othermeans

In conclusion we present a relatively easy method forassessing presumably neurotoxic changes in the rat brain usingT2 parametric mapping and propose that quantification of thesechanges using individual baseline subtraction is the method ofchoice Acquisition and calculation of T2 maps are simple andstraightforward processes much less prone to motion and sus-ceptibility artifacts than diffusion-weighted or diffusion tensorimaging methods and capable of collecting complete 3D imagesof the rat brain in as short a time as 10 min

FUNDING

This work was supported by the National Center forToxicological Research (NCTR) and Center for DrugEvaluation and Research (CDER) US FDA [protocol numberE07418]

ACKNOWLEDGMENTS

This document has been reviewed in accordance with U SFDA policy and approved for publication Approval does notsignify that the contents necessarily reflect the position oropinions of the FDA nor does mention of trade names orcommercial products constitute endorsement or recom-mendation for use The findings and conclusions in thisreport are those of the authors and do not necessarily repre-sent the views of the FDA

REFERENCESAcosta M T Munashinge J Zhang L Guerron D A

Vortmeyer A and Theodore W H (2012) Isolated seizuresin rats do not cause neuronal injury Acta Neurol Scand 12530ndash37

Airaksinen A M Hekmatyar S K Jerome N Niskanen J PHuttunen J K Pitkanen A Kauppinen R A and Grohn OH (2012) Simultaneous BOLD fMRI and local field potentialmeasurements during kainic acid-induced seizures Epilepsia53 1245ndash1253

Airaksinen A M Niskanen J P Chamberlain R HuttunenJ K Nissinen J Garwood M Pitkanen A and Grohn O(2010) Simultaneous fMRI and local field potential measure-ments during epileptic seizures in medetomidine-sedatedrats using raser pulse sequence Magn Reson Med 64 1191ndash1199

Altman D G and Bland J M (1994) Diagnostic tests 1Sensitivity and specificity BMJ 308 1552

Benkovic S A OrsquoCallaghan J P and Miller D B (2004)Sensitive indicators of injury reveal hippocampal damage inC57BL6J mice treated with kainic acid in the absence oftonic-clonic seizures Brain Res 1024 59ndash76

Biomarkers Definitions Working Group (2001) Biomarkers andsurrogate endpoints Preferred definitions and conceptualframework Clin Pharmacol Ther 69 89ndash95

Borsook D Hargreaves R and Becerra L (2011) Can functionalmagnetic resonance imaging improve success rates in CNSdrug discovery Expert Opin Drug Discov 6 597ndash617

Cohen J (1988) Statistical Power Analysis for the BehavioralSciences 2nd ed Lawrence Earlbaum Associates Hillsdale NJ

Coley S C Porter D A Calamante F Chong W K andConnelly A (1999) Quantitative MR diffusion mapping andcyclosporine-induced neurotoxicity AJNR Am J Neuroradiol20 1507ndash1510

Costa L G (1996) Biomarker research in neurotoxicology Therole of mechanistic studies to bridge the gap between thelaboratory and epidemiological investigations EnvironHealth Perspect 104(Suppl 1) 55ndash67

Costa L G (1998) Biochemical and molecular neurotoxicologyRelevance to biomarker development neurotoxicity testingand risk assessment Toxicol Lett 102ndash103 417ndash421

DeOlmos J S and Ingram W R (1971) An improved cupric-sil-ver method for impregnation of axonal and terminal degen-eration Brain Res 33 523ndash529

Deoni S C Rutt B K and Peters T M (2003) Rapid combinedT1 and T2 mapping using gradient recalled acquisition in thesteady state Magn Reson Med 49 515ndash526

Ebisu T Rooney W D Graham S H Mancuso A Weiner MW and Maudsley A A (1996) MR spectroscopic imagingand diffusion-weighted MRI for early detection of kainate-in-duced status epilepticus in the rat Magn Reson Med 36821ndash828

Ebisu T Rooney W D Graham S H Weiner M W andMaudsley A A (1994) N-acetylaspartate as an in vivomarker of neuronal viability in kainate-induced status epi-lepticus 1H magnetic resonance spectroscopic imagingJ Cereb Blood Flow Metab 14 373ndash382

Eichler A F Batchelor T T and Henson J W (2007) Diffusionand perfusion imaging in subacute neurotoxicity followinghigh-dose intravenous methotrexate Neuro Oncol 9 373ndash377

Filippi M Charil A Rovaris M Absinta M and Rocca M A(2014) Insights from magnetic resonance imaging HandbClin Neurol 122 115ndash149

Gad S C (2014) Safety and regulatory requirements and chal-lenge for CNS drug development Neurobiol Dis 61 39ndash46

Goodsaid F M and Frueh F W (2007) Questions and answersabout the pilot process for biomarker qualification at theFDA Drug Discov Today Technol 4 9ndash11

Hanig J Paule M G Ramu J Schmued L Konak TChigurupati S Slikker W Jr Sarkar S and Liachenko S(2014) The use of MRI to assist the section selections forclassical pathology assessment of neurotoxicity RegulToxicol Pharmacol 70 641ndash647 doi 101016jyrtph201409010101016jyrtph201409010

8 | TOXICOLOGICAL SCIENCES 2015

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from

Hopkins K J Wang G and Schmued L C (2000) Temporalprogression of kainic acid induced neuronal and myelin de-generation in the rat forebrain Brain Res 864 69ndash80

Hsu Y H Lee W T and Chang C (2007) Multiparametric MRIevaluation of kainic acid-induced neuronal activation in rathippocampus Brain 130(Pt 12) 3124ndash3134

Hunfeld M Pope K J Fitzgibbon S P Willoughby J O andBroberg M (2013) Effects of anesthetic agents on seizure-in-duction with intra-cortical injection of convulsants EpilepsyRes 105 52ndash61

Immonen R J Kharatishvili I Sierra A Einula C PitkanenA and Grohn O H (2008) Manganese enhanced MRI detectsmossy fiber sprouting rather than neurodegeneration gliosisor seizure-activity in the epileptic rat hippocampusNeuroimage 40 1718ndash1730

Kuchar M Z Jerabkova P Brada J Kaspar J Skrivan J BetkaJ and Langmeier M (2006) Magnetic resonance imaging ofthe rat brain after epileptic seizuresndashpreliminary resultsPrague Med Rep 107 242ndash250

Laitinen T Sierra A Pitkanen A and Grohn O (2010)Diffusion tensor MRI of axonal plasticity in the rat hippocam-pus Neuroimage 51 521ndash530

Lee M S McKinney A M Brace J R and Santacruz K (2010)Clinical and imaging features of fludarabine neurotoxicity JNeuroophthalmol 30 37ndash41

Mavroudis G Prior M J Lister T Nolan C C and Ray D E(2006) Neurochemical and oedematous changes in 13-dini-trobenzene-induced astroglial injury in rat brain from a 1H-nuclear magnetic resonance perspective J Neural Transm113 1263ndash1278

Monti M M (2011) Statistical analysis of fMRI time-seriesA critical review of the GLM approach Front HumNeurosci 5 28

Najm I M Wang Y Shedid D Luders H O Ng T C andComair Y G (1998) MRS metabolic markers of seizures andseizure-induced neuronal damage Epilepsia 39 244ndash250

Nakasu Y Nakasu S Kizuki H Uemura S Morikawa SInubushi T and Handa J (1995a) Changes in water diffu-sion of rat limbic system during status epilepticus elicited bykainate Psychiatry Clin Neurosci 49 S228ndashS230

Nakasu Y Nakasu S Morikawa S Uemura S Inubushi Tand Handa J (1995b) Diffusion-weighted MR in experimen-tal sustained seizures elicited with kainic acid AJNR Am JNeuroradiol 16 1185ndash1192

Olney J W (1990) Excitotoxicity An overview Can Dis WklyRep 16(Suppl 1E) 47ndash57 discussion 57ndash58

Peterson R A Gabrielson K L Allan Johnson G Pomper MG Coatney R W and Winkelmann C T (2011) Continuingeducation course 1 Non-invasive imaging as a problem-solving tool and translational biomarker strategy in toxico-logic pathology Toxicol Pathol 39 267ndash272

Pirttila T R Pitkanen A Tuunanen J and Kauppinen R A(2001) Ex vivo MR microimaging of neuronal damage afterkainate-induced status epilepticus in rat Correlation withquantitative histology Magn Reson Med 46 946ndash954

Preece N E Houseman J King M D Weller R O andWilliams S R (2004) Development of vigabatrin-inducedlesions in the rat brain studied by magnetic resonanceimaging histology and immunocytochemistry Synapse 5336ndash43

Ramsey N F Hoogduin H and Jansma J M (2002) FunctionalMRI experiments Acquisition analysis and interpretation ofdata Eur Neuropsychopharmacol 12 517ndash526

Righini A Pierpaoli C Alger J R and Di Chiro G (1994) Brainparenchyma apparent diffusion coefficient alterations asso-ciated with experimental complex partial status epilepticusMagn Reson Imaging 12 865ndash871

Sandoval C Kutscher M Jayabose S and Tenner M (2003)Neurotoxicity of intrathecal methotrexate MR imaging find-ings AJNR Am J Neuroradiol 24 1887ndash1890

Slikker W Jr (1991) Biomarkers of neurotoxicity An overviewBiomed Environ Sci 4 192ndash196

Slikker W Jr and Bowyer J F (2005) Biomarkers of adult anddevelopmental neurotoxicity Toxicol Appl Pharmacol 206255ndash260

Slikker W Jr Crump K S Andersen M E and Bellinger D(1996) Biologically based quantitative risk assessment ofneurotoxicants Fundam Appl Toxicol 29 18ndash30

Slikker W Jr and Gaylor D W (1995) Risk assessment strate-gies for neuroprotective agents Ann N Y Acad Sci 765 198ndash208 discussion 209

Switzer R C III (1991) Strategies for assessing neurotoxicityNeurosci Biobehav Rev 15 89ndash93

Switzer R C III (2000) Application of silver degeneration stainsfor neurotoxicity testing Toxicol Pathol 28 70ndash83

Switzer R C III Lowry-Franssen C and Benkovic S A (2011)Recommended neuroanatomical sampling practices forcomprehensive brain evaluation in nonclinical safety stud-ies Toxicol Pathol 39 73ndash84

Tang P Liachenko S Melick J A and Xu Y (1998) [31P][1H]nuclear magnetic resonance study of mitigating effects ofGYKI 52466 on kainate-induced metabolic impairment inperfused rat cerebrocortical slices Epilepsia 39 577ndash583

van Vliet E A Otte W M Gorter J A Dijkhuizen R M andWadman W J (2014) Longitudinal assessment of blood-brain barrier leakage during epileptogenesis in rats A quan-titative MRI study Neurobiol Dis 63 74ndash84

Wang Y Majors A Najm I Xue M Comair Y Modic M andNg T C (1996) Postictal alteration of sodium content andapparent diffusion coefficient in epileptic rat brain inducedby kainic acid Epilepsia 37 1000ndash1006

Wolf O T Dyakin V Patel A Vadasz C de Leon M JMcEwen B S and Bulloch K (2002) Volumetricstructural magnetic resonance imaging (MRI) of the rathippocampus following kainic acid (KA) treatment Brain Res934 87ndash96

Woodcock J Buckman S Goodsaid F Walton M K andZineh I (2011) Qualifying biomarkers for use in drug devel-opment A US food and drug administration overview ExpertOpin Med Diagn 5 369ndash374

LIACHENKO ET AL | 9

at FDA

Library on June 12 2015

httptoxscioxfordjournalsorgD

ownloaded from