Oc

Ta

b

c

a

ARRAA

KIfRIB

1

pmiPete

donvrsprGrwnd

0d

Neuropsychologia 47 (2009) 2942–2947

Contents lists available at ScienceDirect

Neuropsychologia

journa l homepage: www.e lsev ier .com/ locate /neuropsychologia

bject priming and recognition memory: Dissociable effects in left frontalortex at encoding

om J. Spencer a, Daniela Montaldi b, Qi-Yong Gong c, Neil Roberts c, Andrew R. Mayes b,∗

Department of Psychology, University of Liverpool, Eleanor Rathbone Building, PO Box 147, Liverpool L69 3BS, UKSchool of Psychological Sciences, University of Manchester, Oxford Road, Manchester M13 9PL, UKMagnetic Resonance and Image Analysis Centre, University of Liverpool, PO Box 147, Liverpool L69 3BX, UK

r t i c l e i n f o

rticle history:eceived 6 July 2008eceived in revised form 19 June 2009ccepted 21 June 2009

a b s t r a c t

Functional magnetic resonance imaging (fMRI) studies have implicated the left prefrontal cortex in prim-ing. We tested the hypothesis that object encoding activity in different prefrontal cortex regions selectivelypredicts subsequent object priming and recognition respectively. Participants were scanned whilst mak-ing semantic category judgements about novel object pictures. One week later priming and recognition

vailable online 27 June 2009

eywords:mplicit memoryMRI

of these objects were tested. Encoding that produced long-lasting priming in the absence of recognitionmemory was associated with increased activity in left inferior prefrontal (BA 47) and superior frontal (BA8) cortices. In contrast, encoding that produced object recognition one week later activated the left middlefrontal cortex (BA 9). This is consistent with other evidence indicating that object priming and recognition

emo

epetition primingnferior prefrontal cortexA47

are independent kind of mare discussed.

. Introduction

Prior processing of an object often facilitates its subsequentrocessing. This phenomenon, called repetition priming, can beeasured behaviourally when subjects are faster at naming or mak-

ng decisions about repeated objects compared with novel objects.riming can occur in the absence of aware memory for the learningpisode (Schacter, Chiu, & Ochsner, 1993) and is therefore thoughto reflect the influence of implicit (unaware) memory for the earlierxposure of the primed stimulus.

Neuroimaging studies of priming have found reduced haemo-ynamic responses in left inferior prefrontal cortex (LIFC) andccipitotemporal regions during processing of repeated relative toovel objects (Simons, Koutstaal, Prince, Wagner, & Schacter, 2003;an Turennout, Ellmore, & Martin, 2000; see Henson, 2003 for aeview). Several studies have investigated how priming relates toubsequent recognition memory with early work suggesting thatriming’s presence during encoding has a negative impact on laterecognition (Wagner, Maril, & Schacter, 2000; although see Stark,ordon, & Stark, 2008). In contrast, little is known about the neu-

al correlates of encoding processes that lead to object priming andhether these processes differ from those that lead to later recog-

ition memory. If priming is the consequence of neural processinguring the first exposure of a stimulus, then encoding activity

∗ Corresponding author. Tel.: +44 161 275 2579; fax: +44 161 275 2623.E-mail address: Andrew.mayes@manchester.ac.uk (A.R. Mayes).

028-3932/$ – see front matter © 2009 Elsevier Ltd. All rights reserved.oi:10.1016/j.neuropsychologia.2009.06.022

ry. Problems of measuring item-by-item recognition and priming together

© 2009 Elsevier Ltd. All rights reserved.

should be able to predict subsequent object priming. This is analo-gous to the ‘subsequent memory’ approach where brain responsesat encoding are associated with later recognition memory (Brewer,Zhao, Desmond, Glover, & Gabrieli, 1998; Wagner et al., 1998).

One candidate brain region that might mediate encoding thatleads to priming is the LIFC. A recent study by Wig, Grafton,Demos, and Kelley (2005) provides evidence that activity in theLIFC during encoding of objects is necessary to produce subsequentbehavioural priming of those objects. In this study, transcra-nial magnetic stimulation was applied to subjects’ LIFC whilstthey made living/non-living judgements about objects. Disrup-tion of neural processing in LIFC at encoding abolished bothlater behavioural priming and repetition-related haemodynamicdecreases in the LIFC. This suggests LIFC activity at encoding isimportant for later object priming. Consistent with this view, Schottet al. (2006) have shown that reduced activity in LIFC is associ-ated with later word-stem completion priming. Their study wasable to distinguish between the encoding of words that were eitherremembered, primed but not remembered, or forgotten (neitherprimed nor remembered). This design was unable to examinewords that were later primed and remembered, i.e., priming in thecontext of recognition.

The present study was designed to test the hypothesis that activ-

ity in the LIFC at encoding predicts subsequent object priming, butthat different frontal activity predicts subsequent object recogni-tion (Brewer et al., 1998). To do this we scanned participants withfMRI whilst they performed semantic category judgements aboutnovel object pictures. One week later participants completed a

ycholo

ptauptam2te

2

2

i2ia

2

rw4arfwapwbb

iatcstftitsaatepa2

2

kba(aw

2

Druinasb

T.J. Spencer et al. / Neurops

riming task followed by a surprise recognition test. This allowed uso classify each object at encoding as later primed or later unprimed,nd also as later recognized or unrecognized. This design alloweds to examine the neural correlates of encoding that lead to laterriming and compare them with correlates of encoding that leado recognition. Direct measures of priming (reaction time (RT))nd recognition were made for each trial, rather than calculatingeans across subjects as other studies have done (Wagner et al.,

000). This approach is novel and allowed encoding effects leadingo priming in the context of recognition or non-recognition to bexamined separately.

. Methods

.1. Subjects

Twelve right-handed participants with no history of neurological or psychiatricllness gave informed consent to participate in the study (nine women; mean age2.6 years; age range 19–34 years). All the participants underwent a medical exam-

nation before taking part in the study, which had received local ethics committeepproval.

.2. Stimuli and experimental procedure

Stimuli consisted of 240 black and white line drawings of common objects (120epresenting man-made objects and 120 representing natural objects). 200 objectsere taken from a standardised set (Snodgrass & Vanderwart, 1980) and a further

0 natural objects were drawn in the same style. Pictures were counterbalancedcross participants so that, at test, each picture was novel for some participants andepeated for the others. During the study phase, participants were scanned usingMRI whilst they saw a series of 160 object pictures. The subjects’ task was to decide

hether each picture represented a man-made or a natural object as quickly ands accurately as possible by pressing one of two response buttons. Pictures wereresented every 6 s (2 s presentation, 4 s inter-stimulus interval) using Psyscope soft-are (Cohen, MacWhinney, Flatt, & Provost, 1993). RTs were recorded from response

oxes connected to a Psyscope button box and a Power Mac G3 computer. Responseuttons were counterbalanced across subjects.

One week later the same participants returned for the test phase of the exper-ment and were scanned whilst performing two tasks; a priming task followed by

surprise recognition test. In the priming task, participants saw all the object pic-ures presented at study plus 80 novel objects randomly intermixed. They againategorised each one as man-made or natural as quickly and as accurately as pos-ible. During the recognition test, participants saw all the object pictures again, buthis time they were asked to decide whether each object was the one they recognizedrom the study phase. Participants were also strongly instructed only to respond thathey did not recognize an object if they were very confident that they had not seent one week earlier. In order to ensure that familiarity was used as well as recollec-ion (source memory), participants were told to respond ‘recognized’ even withoutource memory, provided familiarity was sufficiently strong. Stimuli for the primingnd recognition tasks were presented using the same parameters as at study (seebove). Objects were presented in the same order during the priming and recogni-ion tests to guarantee that the delays between priming and recognition tests forach object were equal, minimising noise in the recognition measure. The study,riming and test phases were split into 8-min sessions (two at study and three eacht priming and recognition) with 80 object pictures presented in each session and a-min interval between each session.

.3. Image acquisition

A 1.5T LX/Nvi Neuro-Optimised MR imaging system (General Electric, Milwau-ee, USA) was used to acquire T2* weighted gradient-echo echoplanar images withlood oxygenation level dependent (BOLD) contrast. A total of 384 volumes werecquired at study (192 per session). Each volume comprised 20, 6 mm axial slices64 × 64, 3 mm × 3 mm pixels, echo time = 40 ms) oriented parallel to the AC-PC linend positioned to cover the entire brain except the most superior part. Volumesere collected continuously with a repetition time of 2.5 s/volume.

.4. fMRI analysis

Data were analysed using Statistical Parametric Mapping (SPM99, Wellcomeepartment of Cognitive Neurology, London, UK). The time series for each voxel was

ealigned temporally to acquisition of the first slice and spatially to the first volume

sing sinc interpolation. Images were normalised to a standard EPI template based

n Talairach space (Ashburner & Friston, 1999) using nonlinear basis functions. Theormalised images were smoothed with an isotropic 8 mm FWHM Gaussian kernelnd proportionally scaled to a grand mean of 100 across all voxels and scans within aession. The time series were high-pass filtered to 1/120 Hz and low-pass smoothedy a 4-s FWHM Gaussian kernel.

gia 47 (2009) 2942–2947 2943

Statistical analysis was carried out in two stages. In the first stage, the BOLDresponse to each event-type was modelled by convolving a series of delta functionscorresponding to each stimulus onset with the canonical haemodynamic responsefunction and its partial derivative with respect to time (Friston et al., 1998). Thesefunctions, together with a constant term for each session, were modelled as partici-pant specific covariates in a fixed effects general linear model. Parameter estimateswere calculated for each covariate from a least mean squares fit of the model to thedata. Pairwise contrasts between the canonical parameter estimates for event-typesgenerated statistical parametric maps (SPMs) of the t-statistic, which were subse-quently transformed into maps of the Z statistic. Contrast images generated for eachparticipant were used in second stage, one-sample t-tests treating participants as arandom effect.

Events of interest in the study phase were defined for objects correctly classi-fied as man-made or natural; incorrect responses were modelled separately. Eachobject was classified as recognized or unrecognized during the recognition test, andthen classified as primed or unprimed depending on whether it was categorised asman-made or natural faster at test than at study (‘inclusive’ analysis condition). Aconsequence of using this definition of priming was that primed items had longerRTs at study than unprimed items. Any speed-up from study to test could thusreflect either priming or regression towards the mean. To minimize effects of regres-sion to the mean an unbiased selection procedure was used to ensure that primedand unprimed objects had matched RT distributions at study (‘matched’ analysiscondition). The rationale is that regression to the mean should operate equally forprimed and unprimed stimuli with matched study RTs. Use of the ‘matched’ anal-ysis condition also ensured that encoding was more similar between later primedand unprimed stimuli. An algorithm written in Matlab (version 5.3, Mathworks) tookprimed and unprimed response distributions for each subject and binned them every50–70 ms based on their study RT. For corresponding pairs of primed and unprimedbins the minimum number of trials in either bin was selected. A matched number oftrials was then randomly selected from the corresponding bin of the other distribu-tion. This procedure was conducted separately for both recognized and unrecognizedobjects. The resultant trials had matching RT distributions and did not differ in termsof study median RTs. Study RT matched and non-matched objects were modelledseparately in the design matrix. Even if regression to the mean is not fully ruled outby this method, regression should only add noise to the data and reduce the chanceof identifying real encoding effects that predict priming (type 2 error), but shouldnot produce false positive fMRI findings. Convergent findings from the ‘inclusive’ and‘matched’ analyses predicting priming can be treated with confidence. The inclusionanalysis should be more subject to noise from stronger regression effects, whereasmatching will have reduced numbers because of the selection process.

In the ‘inclusive’ analysis, second-level contrast images were analysed in a2(primed/unprimed) × 2(recognized/unrecognized) ANOVA to assess main effectsof recognition and priming at encoding as well as any interaction between recog-nition and priming. To further assess encoding effects predicting later recognitionof objects, second level correlations were performed between contrast images forthe Recognized > Unrecognized contrast and subjects’ proportion of hits minus pro-portion of false alarms score (pHits − pFA). To examine the neural correlates ofencoding that lead to priming without recognition, planned comparisons betweenunrecognized primed and unrecognized unprimed second-level contrast imageswere performed. Two versions of this contrast were conducted. The first combinedmatched and unmatched unrecognized objects, and the second used only unrecog-nized objects with matched study RTs to control for regression towards the mean.Similar comparisons between recognized primed and recognized unprimed objectswere also performed. Regions in SPMs were considered significant if five or morecontiguous voxels survived a P < 0.001 uncorrected threshold. There is a risk of mak-ing type I errors with uncorrected thresholds; however, given that this is the firststudy to examine neural correlates of visual object priming at encoding, specifyinga precise a priori anatomical search area to correct for multiple comparisons proveddifficult. The approach adopted here is a reasonable trade-off between making typeI and type II errors (see Schott et al., 2006).

3. Results

3.1. Behavioural results

Performance was close to perfect on the man-made/naturaltask (96% accuracy, SEM 0.4%). Objects with RTs less than250 ms or greater than 2.5 standard deviations above the sub-ject’s mean categorisation RT were considered outliers andwere not analysed. Performance on the recognition test wassignificantly above chance, the group mean corrected recog-nition score (proportion recognized − proportion false alarms)

was 0.28 [t(11) = 8.61, P < 0.001]. Across the group of 12 partici-pants, repeated objects were categorised more quickly (675 ms)than novel objects (692 ms) [F(1,11) = 13.62, P < 0.005] showinga significant group priming effect of 17 ms. Significant primingeffects were also found when recognized (678 ms) and unrec-

2 ycholo

ooByrloopStnbP

tsrttuSbuRpuiyr1

3

w

FobE

944 T.J. Spencer et al. / Neurops

gnized (669 ms) objects were separately compared to novelbjects at test [F(1,11) = 2.91, P < 0.05; F(1,11) = 4.21, P < 0.005].efore matching study RTs, mean correct RTs were anal-sed using a 2(primed/unprimed) × 2(recognized/unrecognized)epeated measures ANOVA at both study and test. At study, RTs wereonger to subsequently primed objects (774 ms) than unprimedbjects (622 ms) [F(1,11) = 195.43, P < 0.0001]. There was no effectf recognition [F(1,11) = 0.06, P > 0.8] and the interaction betweenriming and recognition was not significant [F(1,11) = 0.08, P > 0.79].imilarly, at test, RTs were longer to unprimed objects (758 ms)han primed objects (617 ms) [F(1,11) = 41.88, P < 0.0001]. There waso effect of recognition [F(1,11) = 1.36, P > 0.26] and the interactionetween priming and recognition was not significant [F(1,11) = 0.65,> 0.43].

After matching primed and unprimed objects for study RTs,here was no significant difference between median study RTs forubsequently primed recognized pictures (649 ms) and unprimedecognized pictures (648 ms) [t(11) = 0.287, P > 0.77] indicating thathe matching procedure was successful. However, median RTs atest were shorter for primed recognized objects (572 ms) thannprimed recognized objects (778 ms) [t(11) = −7.28, P < 0.001].imilarly, for unrecognized pictures there was no differenceetween median study RTs for subsequently primed (659 ms) andnprimed (647 ms) objects [t(11) = 1.82, P < 0.09] when encodingTs were matched. However, median RTs at test were shorter forrimed unrecognized objects (581 ms) than they were for unprimednrecognized objects (754 ms) [t(11) = 8.93, P < 0.0001]. To give an

ndication of the number of trials that went into the priming anal-sis, there were on average: 59.7 recognized primed items, 43.6ecognized unprimed items, 24.4 unrecognized primed items and8.6 unrecognized unprimed items per participant.

.2. fMRI results

The main effect of recognition showed that encoding responsesere increased for later recognized objects relative to later unrec-

ig. 1. Brain activity in left middle frontal gyrus and right fusiform gyrus at encoding was abjects were found in left middle frontal gyrus (A, 7 voxels) and right fusiform gyrus (B, 21rain. Graphs show plots of peak percentage signal change (relative to the mean over all voxrror bars show the standard error of the mean across subjects.

gia 47 (2009) 2942–2947

ognized objects in left middle frontal gyrus, BA 9 (Fig. 1A;x = −33, y = +6, z = +39, Z-score = 3.52), right fusiform gyrus (Fig. 1B;x = +48, y = −54, z = −12, Z-score = 3.97) and left inferior temporalgyrus (x = −42, y = −66, z = −9, Z-score = 3.81). In the reverse com-parison (Unrecognized > Recognized) no significant voxels werefound. Positive correlations between Recognized > UnrecognizedBOLD responses at encoding and participants’ corrected recogni-tion scores revealed significant effects in right parahippocampalgyrus (x = +21, y = −36, z = −9, Z-score = 3.29), inferior parietal cor-tex (x = −51, y = −30, z = +48, Z-score = 3.33) and superior parietalcortex (x = −33, y = −60, z = +57, Z-score = 4.02).

The main effect of priming revealed significantly greater acti-vation for later-primed relative to later-unprimed items in leftinferior frontal gyrus (x = −42, y = +21, z = −3, Z-score = 4.18), supe-rior frontal gyrus (x = +3, y = +27, z = +42, Z-score = 3.35) and bilateralcaudate nucleus (x = −9, y = +15, z = 0, Z-score = 3.53; x = +12, y = +3,z = +15, Z-score = 3.55). No significant voxels were found in thereverse comparison (Unprimed–Primed) and there were no signif-icant interactions between priming and recognition.

To assess encoding effects that lead to priming in the absenceof recognition, responses to primed and unprimed unrecognizedobjects were compared. The first analysis included all unrecog-nized items, combining objects with matched and unmatched RTsat study (‘inclusive’); the second ensured that study RTs werematched for primed and unprimed objects (‘matched’). The ‘inclu-sive’ comparison revealed greater encoding activation for primedunrecognized items relative to unprimed unrecognized items inbilateral inferior prefrontal cortex (Fig. 2A; x = −42, y = +21, z = −12,Z-score = 3.57; x = +36, y = +24, z = −9, Z-score = 3.48), bilateral headof caudate nucleus (left: x = −9, y = +12, z = +3, Z-score = 3.50; right:x = +12, y = +9, z = +3, Z-score = 3.91) and left body of the cingulate

gyrus (x = −15, y = −3, z = +45, Z-score = 4.47). Two regions in BA19 showed decreased encoding responses for later unrecognizedprimed objects compared with unrecognized unprimed objects(left cuneus: x = −9, y = −93, z = +27, Z-score = 4.12; right precuneus:x = +18, y = −84, z = +42, Z-score = 4.02).

ssociated with later recognition. Greater responses to recognized than unrecognizedvoxels). Significant effects are rendered on coronal slices of a canonical normalised

els) for subsequently primed and unprimed objects as a function of later recognition.

T.J. Spencer et al. / Neuropsychologia 47 (2009) 2942–2947 2945

Fig. 2. Brain activity in left inferior frontal cortex at encoding was associated with later priming. Greater responses to primed than unprimed unrecognized objects werefound in left inferior prefrontal cortex (A, 60 voxels) when the analysis used all unrecognized trials unmatched for study RTs (‘inclusive’ analysis). Similar effects were foundwhen unrecognized trials were restricted to primed and unprimed objects with matched study RTs (‘matched’ analysis; B, 16 voxels). The matched analysis also revealed aregion in left superior frontal cortex that predicted priming in the absence of recognition (C, 23 voxels). Significant effects are rendered on coronal and sagittal slices of ac

iszpwom

uerfnefo

anonical normalised brain. See Fig. 1 legend for further details.

The ‘matched’ comparison revealed increased encoding activityn left inferior prefrontal cortex (Fig. 2B; x = −51, y = +21, z = 0, Z-core = 3.51) and left superior frontal cortex (Fig. 2C; x = −12, y = +30,= +48, Z-score = 3.43) for later unrecognized primed objects com-ared with unrecognized unprimed objects. No significant voxelsere found showing reduced activity for unrecognized primed

bjects compared with unrecognized unprimed objects in theatched analysis.

Both ‘inclusive’ and ‘matched’ comparisons between primed andnprimed recognized objects revealed no significant voxels. How-ver, there were no significant interactions between priming andecognition and plots of parameter estimates for the left inferior

rontal region predicting priming appear similar for both recog-ized and unrecognized items (see Fig. 2). This suggests similarncoding effects predicting priming in the left inferior frontal gyrusor recognized and unrecognized stimuli, although the effect wasnly significant for the unrecognized objects.

4. Discussion

The main finding in the present study was that activation ofdifferent regions in left frontal cortex during encoding of object pic-tures predicted subsequent priming and subsequent recognition.Specifically, increased responses in the inferior aspect of prefrontalcortex and left superior frontal cortex predicted later priming with-out recognition memory. In contrast, increased responses in leftmiddle frontal gyrus were associated with later recognition of theobjects. Encoding activations predicting later priming were found incomparisons between unrecognized primed and unprimed objectswith matched study RTs, so they were unrelated to time on task

differences at encoding. These findings provide evidence that dif-ferent frontal encoding activations predict later object priming andrecognition respectively.

Several neuroimaging studies have demonstrated decreasedactivity in the LIFC during priming (van Turennout et al., 2000;

2 ycholo

vTpitWprVfipntstSKe

daati(isdb(foTlosdhppottpudatdorS

lrTlpdeepiups

946 T.J. Spencer et al. / Neurops

an Turennout, Bielamowicz, & Martin, 2003; Wig et al., 2005).hese effects have been linked to conceptual priming, rather thanerceptual priming, because repetition-related response decreases

n the LIFC appear to be independent of the perceptual form ofhe primed stimuli (Vuilleumier, Henson, Driver, & Dolan, 2002;

agner, Desmond, Demb, Glover, & Gabrieli, 1997). Furthermore,riming deactivations in this region have been shown to cor-elate with behavioural measures of priming (Dobbins, Schnyer,erfaellie, & Schacter, 2004; Maccotta & Buckner, 2004). Thesendings are consistent with the idea that, during priming, partici-ants access previously retrieved information more fluently thanovel information, and this requires less dependence on execu-ive control processes in LIFC (Badre & Wagner, 2002). Althoughome authors have claimed that lesions of the left frontal cor-ex do not impair priming (e.g. word-stem completion priming,himamura, Gershberg, Jurica, Mangels, & Knight, 1992), Swick andnight (1996) have shown evidence that left frontal cortex is nec-ssary for normal semantic-decision priming.

Recent research has suggested that visual object priming may beependent on learning stimulus–response associations (Dobbins etl., 2004). This view proposes that when a stimulus is first presentedn association is set up between that stimulus and the response itriggers. By reactivating these stimulus–response associations dur-ng priming many processing steps may be bypassed and fasterprimed) responses are produced to these stimuli. By simply chang-ng the required response to object pictures from ‘bigger than ahoebox’ at first presentation of objects, to ‘smaller than a shoebox’uring their second presentation, Dobbins et al. showed that bothehavioural priming and neural response suppression was reducedDobbins et al., 2004). Further work has suggested that the pre-rontal cortex, including the LIFC, might be sensitive to retrievalf these stimulus–response associations (Horner & Henson, 2008).hus, greater, and perhaps more efficient semantic and/or phono-

ogical processing at encoding in the LIFC may lead to the formationf a more effective stimulus–response association memory repre-entation and reactivation of this produces a more fluent responseuring the reprocessing of that stimulus. A direct prediction of thisypothesis is that the same LIFC regions will deactivate during laterriming. Unpublished data from our laboratory has revealed objectriming deactivations at test in similar LIFC regions (new unrec-gnized unprimed > unrecognized primed: x = −48, y = +15, z = 0) tohe encoding activations that predicted priming without recogni-ion memory (x = −51, y = +21, z = 0). LIFC effects at encoding andriming did not overlap exactly when masking procedures weresed to assess overlap. Despite this, it seems likely that these effectserive from the same functional area of left frontal cortex, given themount of noise in the spatial registration process and the fact thathese regions were separated by only a few voxels. Similar retrievaleactivations have been found in association with familiarity mem-ry but at sites, such as the perirhinal cortex, that are probably notelated to priming (Davachi, Mitchell, & Wagner, 2003; Montaldi,pencer, Roberts, & Mayes, 2006).

The view that activity in LIFC during initial exposure of a stimu-us determines later behavioural and neural priming in the sameegion is supported by recent findings by Schott et al. (2006).hey found that activity in bilateral inferior frontal cortex and

eft fusiform not only predicted subsequent word-stem completionriming, but also that these regions showed decreased responsesuring priming without recognition. However, their encodingffects in these brain areas were in the opposite direction to ourncoding effects; i.e., they found decreased responses for later

rimed words compared to later unprimed words. Discrepancies

n the results of the two studies could reflect different mechanismsnderlying word-stem completion priming and semantic-decisionriming of objects, or could be due to methodological differences,uch as how regression towards the mean was controlled in our

gia 47 (2009) 2942–2947

study. It is interesting to note that Schott et al. (2006) did iden-tify other regions in frontal and parietal cortices that showedincreased encoding activations predictive of later priming versusnon-priming. These effects, however, were not associated withpriming deactivations at retrieval. Further work is needed to clar-ify which brain regions and processes at encoding are necessary toproduce later priming.

Encoding effects that predicted later priming were also foundin left superior frontal gyrus (∼BA 8). This region has been lesswell studied than lateral prefrontal cortex, but is thought tointeract with LIFC during semantic processing (Demonet et al.,1992) and has been shown to be more active when subjectsmake semantic judgments compared with non-semantic judg-ments about words (Demb et al., 1995; Gabrieli et al., 1996). Thesestudies found decreases in activation in left superior frontal gyruswith repeated performance on a semantic word classification task.Further evidence that this region is involved in priming comesfrom an experiment by Wagner et al. (1997) that used a similarcategorisation task to the one used in the present study (living/non-living decision). They found priming deactivations in LIFC and leftsuperior frontal gyrus associated with both picture and word rep-etition.

Although no activation differences were found between primedand unprimed recognized objects, plots of parameter estimates inLIFC and superior frontal cortex revealed a similar pattern of activ-ity to primed and unprimed unrecognized objects in these regions(Fig. 2). Priming with recognition may well have been noisier thanpriming without recognition in the current experiment becauseparticipants were instructed to classify stimuli as recognized evenif they could not recollect their source, provided the stimuli feltsufficiently familiar. We suggest that encoding activations at thesesites are related to the production of priming and are independentof the production of recognition.

All event-related fMRI priming studies in which RT is the mea-sure of priming need to measure recognition and priming for eachstimulus. Accurate measurement of both priming and recogni-tion on an item-by-item basis leads to unavoidable problems. Inour study, the recognition test was performed after the experi-ment’s second semantic classification (priming) phase. Althoughthe delayed recognition testing would have reduced interferencewith priming, fluctuations of memory over time would have slightlyreduced the accuracy with which recognition around the time ofpriming was measured. So, some stimuli would have been recog-nized around the time of priming, but not at the final test (andvice-versa). Given that the recognition of interest was at the timeof priming, the delayed measure of recognition may have involvedmore source memory to decide whether objects had only been seenduring the priming test or also during the study phase. There was,therefore, a risk that our participants may have used recollectionmore than they would have if tested around the time of priming, andso placed less reliance on familiarity. The greater need for sourcememory may have led to objects being judged as unrecognized evenif they had been incidentally recognized during the priming phase.We reduced these confounds in two main ways. First, the longretention delay made the temporal source discrimination easierbecause the priming test-recognition test delay was only minuteslong whereas the study-priming test delay was one week. Second,we stringently instructed participants to classify stimuli as recog-nized even if they could not recollect their source clearly, providedthe stimuli felt sufficiently familiar. Our aim was to minimize thechance that objects classified as unrecognized would have been rec-

ognized as familiar during the priming test. But future research mayneed to check this directly by measuring recollection and famil-iarity strength of stimuli. This would also allow recognition to bequantitatively as well as qualitatively matched between primed andunprimed stimuli.

ycholo

tdmupvttaa(mbaatasoc

R

A

B

B

C

D

D

D

D

T.J. Spencer et al. / Neurops

The difficulty of assessing RT-based priming as well as recogni-ion item-by-item at the time of priming can be addressed as weid in which case recognition involves a source component thatakes it harder to be confident that unrecognized items are both

nrecollected and unfamiliar. Testing recognition before assessingriming is unacceptable because stimulus recognition testing pro-ides more encoding that affects RTs during the subsequent primingask. With non-RT-based forms of priming, such as stem comple-ion, recall can be tested immediately before priming, but this onlyllows analysis of priming without explicit memory (see Schott etl., 2006). Another option would be to follow every classificationpriming) judgement with a recollection/familiarity (RF) judge-

ent (i.e., on an item-by-item basis), but this is also problematicecause participants’ performance on the priming task is unavoid-bly influenced by their subsequent RF decision even though theyre told to focus first only on the classification and then only onhe RF task. To increase confidence in results, future research mightttempt to run parallel studies using the delayed recognition along-ide item-by-item priming and recognition testing procedures,nly accepting findings when results from the two proceduresonverge.

eferences

shburner, J., & Friston, K. (1999). Nonlinear spatial normalization using basis func-tions. Human Brain Mapping, 7, 254–266.

adre, D., & Wagner, A. D. (2002). Semantic retrieval, mnemonic control, and pre-frontal cortex. Behavioral and Cognitive Neuroscience Reviews, 1, 206–218.

rewer, J. B., Zhao, Z., Desmond, J. E., Glover, G. H., & Gabrieli, J. D. (1998). Mak-ing memories: Brain activity that predicts how well visual experience will beremembered. Science, 281, 1185–1187.

ohen, J. D., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope: A newgraphic interactive environment for designing psychology experiments. BehaviorResearch Methods, Instruments, & Computers, 25, 257–271.

avachi, L., Mitchell, J. P., & Wagner, A. D. (2003). Multiple routes to memory: Distinctmedial temporal lobe processes build item and source memories. Proceedings ofthe National Academy of Sciences USA, 100, 2157–2162.

emb, J. B., Desmond, J. E., Wagner, A. D., Vaidya, C. J., Glover, G. H., & Gabrieli,J. D. E. (1995). Semantic encoding and retrieval in the left inferior prefrontalcortex a functional MRI study of task difficulty and process specificity. Journal ofNeuroscience, 15, 5870–5878.

emonet, J. F., Chollet, F., Ramsay, S., Cardebat, D., Nespoulous, J. L., Wise, R., et al.(1992). The anatomy of phonological and semantic processing in normal sub-jects. Brain, 115, 1753–1768.

obbins, I. G., Schnyer, D. M., Verfaellie, M., & Schacter, D. L. (2004). Cortical activityreductions during repetition priming can result from rapid response learning.Nature, 428, 316–319.

gia 47 (2009) 2942–2947 2947

Friston, K. J., Fletcher, P., Josephs, O., Holmes, A. P., Rugg, M. D., & Turner, R. (1998).Event-related fMRI: Characterizing differential responses. NeuroImage, 7, 30–40.

Gabrieli, J. D. E., Desmond, J. E., Demb, J. B., Wagner, A. D., Stone, M. V.,Vaidya, C. J., et al. (1996). Functional magnetic resonance imaging of seman-tic memory processes in the frontal lobes. Psychological Science, 7, 278–283.

Henson, R. N. A. (2003). Neuroimaging studies of priming. Progress in Neurobiology,70, 53–81.

Horner, A. J., & Henson, R. N. (2008). Priming, response learning and repetitionsuppression. Neuropsychologia, 46, 1979–1991.

Maccotta, L., & Buckner, R. L. (2004). Evidence for neural effects of repetition thatdirectly correlate with behavioral priming. Journal of Cognitive Neuroscience, 16,1625–1632.

Montaldi, D., Spencer, T. J., Roberts, R., & Mayes, A. R. (2006). The neural system thatmediates familiarity memory. Hippocampus, 16, 504–520.

Schacter, D. L., Chiu, C. Y. P., & Ochsner, K. N. (1993). Implicit memory: A selectivereview. Annual Review of Neuroscience, 16, 159–182.

Schott, B. H., Richardson-Klavehn, A., Henson, R. N. A., Becker, C., Heinze, H.-J., &Düzel, E. (2006). Neuroanatomical dissociation of encoding processes related topriming and explicit memory. Journal of Neuroscience, 26, 792–800.

Shimamura, A. P., Gershberg, F. B., Jurica, P. J., Mangels, J. A., & Knight, R. T. (1992).Intact implicit memory in patients with frontal lobe lesions. Neuropsychologia,30, 931–937.

Simons, J. S., Koutstaal, W., Prince, S., Wagner, A. D., & Schacter, D. L. (2003). Neu-ral mechanisms of visual object priming: Evidence for perceptual and semanticdistinctions in fusiform cortex. NeuroImage, 19, 613–626.

Snodgrass, J. G., & Vanderwart, M. (1980). A standardised set of 260 pictures: Normsfor name agreement, familiarity, and visual complexity. Journal of ExperimentalPsychology: Human Learning and Memory, 6, 174–215.

Stark, S. M., Gordon, B., & Stark, C. E. L. (2008). Does the presence of priming hindersubsequent recognition or recall performance. Memory, 16, 157–173.

Swick, D., & Knight, R. T. (1996). Is prefrontal cortex involved in cued recall? Aneuropsychological test of PET findings. Neuropsychologia, 34, 1019–1028.

van Turennout, M., Bielamowicz, L., & Martin, A. (2003). Modulation of neural activityduring object naming: Effects of time and practice. Cerebral Cortex, 13, 381–391.

van Turennout, M., Ellmore, T., & Martin, A. (2000). Long-lasting cortical plasticityin the object naming system. Nature Neuroscience, 3, 1329–1334.

Vuilleumier, P., Henson, R. N., Driver, J., & Dolan, R. J. (2002). Multiple levels of visualobject constancy revealed by event-related fMRI of repetition priming. NatureNeuroscience, 5, 491–499.

Wagner, A. D., Desmond, J. E., Demb, J. B., Glover, G. H., & Gabrieli, J. D. E. (1997).Semantic repetition priming for verbal and pictorial knowledge: A functionalMRI study of left inferior prefrontal cortex. Journal of Cognitive Neuroscience, 9,714–726.

Wagner, A. D., Maril, A., & Schacter, D. L. (2000). Interactions between forms ofmemory: When priming hinders new episodic learning. Journal of CognitiveNeuroscience, 12, 52–60.

Wagner, A. D., Schacter, D. L., Rotte, M., Koutstaal, W., Maril, A., Dale, A. M., et al.(1998). Building memories: Remembering and forgetting of verbal experiencesas predicted by brain activity. Science, 281, 1188–1191.

Wig, G. S., Grafton, S. T., Demos, K. E., & Kelley, W. M. (2005). Reductions in neuralactivity underlie behavioral components of repetition priming. Nature Neuro-science, 9, 1228–1233.