A repetition suppression effect lasting several days within the semantic network
- First Online:
- Cite this article as:
- Meister, I.G., Buelte, D., Sparing, R. et al. Exp Brain Res (2007) 183: 371. doi:10.1007/s00221-007-1051-8
- 89 Views
Performance in a semantic task is speeded up for repeated stimuli compared to novel stimuli. This conceptual priming effect is related to a decrease in functional activation within the left inferior prefrontal cortex for repeated stimulus exposure (repetition suppression). However, in contrast to perceptual priming which is known to be very robust over long periods of time, previous studies on semantic priming focused on short-term effects. The present study combined a behavioral and functional imaging experiment to investigate long-term conceptual repetition priming (retention interval 3 days). We found a highly significant decrease of reaction time for word stimuli which were presented repeatedly after 3 days both compared to initial presentation and to a matched word list. The functional magnetic resonance imaging data showed a repetition suppression within the left inferior (BA45, BA47) and middle (BA9) frontal gyrus for the comparison of known with unknown words. These data demonstrate that even over a period as long as 3 days, a repetition suppression within the left frontal network involved in semantic decision can be found. Thus, priming-related mechanisms in the semantic network may be robust over several days.
KeywordsFunctional imagingPrimingSemanticPsychophysicalImplicit memoryfMRI
Priming is an implicit form of memory denoting changes in task performance as a result of previous encounter with an item (Tulving and Schacter 1990). According to the underlying main task demands, a differentiation in perceptual and conceptual forms of priming has been proposed (Roediger and McDermott 1993). The priming effect is assumed to be independent of explicit memory (Schott et al. 2005, 2006), rather reflecting enhanced effectivity of cortical stimulus representations (Grill-Spector et al. 2006). There is evidence from neuroimaging studies (Henson 2003) and TMS studies (Wig et al. 2005) that the behavioral phenomenon of repetition priming is causally related to a decrease in functional magnetic resonance imaging (fMRI) activation within task-critical cortical networks.
A remarkable feature of the perceptual priming effect is its longevity: behavioral studies investigating Picture priming (Mitchell and Brown 1988; Cave 1997), repetition priming in reading (Kolers 1976) and fragment priming (Sloman et al. 1988) have demonstrated significant behavioral effects over the course of months, whereas recognition performance declined rapidly. The decline of explicit memory performance is pronounced several days after initial exposure to the stimulus, whereas implicit memory tested with the same retention intervals is relatively stable (Gardiner and Java 1991; Goshen-Gottstein and Kempinsky 2001; Tunney 2003). Recent functional imaging studies accordingly found significant reductions in fMRI activity (repetition suppression) for picture priming over retention intervals of 3 days (van Turennout et al. 2000) and 6 weeks between first and second stimulus encounter (Meister et al. 2005).
The present study sought to extend these previous results using a semantic decision task to investigate long-term priming effects. Whereas it is undisputed that perceptual priming lasts over very long time spans, evidence for the semantic system is very limited. Several recent reports found reductions in fMRI activation within the left inferior frontal cortex when different semantic tasks are performed on repeated stimuli (Demb et al. 1995; Wagner et al. 2000; Donaldson et al. 2001; Copland et al. 2003). It has been shown that the semantic priming effect spans at least many intervening items (Becker et al. 1997; Joordens and Becker 1997); but retention intervals exceeding several minutes have not been reported. The present study combined a behavioral and a fMRI experiment to investigate a retention interval much longer than used in previous studies, investigating if implicit memory in the semantic system shows a longevity comparable to perceptual priming. Healthy subjects were asked to perform abstract/concrete decisions on German nouns comparing novel words and words which were repeated after a retention interval of 3 days. We hypothesized that there would be priming-related decreases of reaction time (RT) in the behavioral paradigm and a corresponding reduction of fMRI activation in the left inferior prefrontal cortex (LIPC), which plays a crucial role for semantic categorisation (Bokde et al. 2001; Wagner et al. 1997; McDermott et al. 2003; Vigneau et al. 2006).
A total of 22 participants (mean age 23.9 years, 12 males) participated in the study (12 of these subjects took part in the fMRI study, mean age 23.1 years, 6 males). The study was approved by the local ethics committee and all participants gave their written informed consent.
Three days prior to the experiment subjects were presented 70 German nouns (35 concrete, 35 abstract: baseline set), which had to be classified as abstract or concrete [the words were classified and grouped according to the ratings of Hager and Hasselhorn (1994)]. The words were centrally presented on a computer screen for 500 ms with an inter-stimulus interval of 2.3 s. Three days later, this baseline set of words was presented randomly intermixed with the set of novel words comprising 70 words, as well, which were matched with the baseline set regarding length, frequency and challenge. Every stimulus was only presented once in a given session. Subjects were asked to perform judgement for abstract or concrete classification as quick as possible by pressing left/right mousebuttons with the right hand. For statistical analysis, averaged individual RTs and percentage of correct responses were used; statistical comparison was done using pairwise paired Student’s t tests. As in this setup every data set was used twice for statistical comparison, a Bonferroni correction was applied to the results (corrected significance level: P < 0.025).
For the fMRI study, we employed a fast event-related design using inter-stimulus intervals randomized between 2.32 and 2.4 s using the task described above. As baseline, subjects fixated of a central hair cross. Presentation was done via high-resolution MRI compatible 3D glasses (Resonance Technology, Van Nuys, CA, USA) in pseudorandom order for 200 ms each. The participants were instructed to perform abstract/concrete judgements as fast as possible using MRI-compatible push-buttons.
A total of 182 functional MR images were acquired on a Philips Gyroscan 1.5 T scanner using a gradient echo EPI sequence (TR = 2.6 s, TE = 50 ms, flip angle = 90°, slice thickness = 4.2 mm, image matrix = 64 × 64). A total of 28 axial slices covering the whole brain were acquired in bottom-to-up slice order.
The fMRI data from each subject were processed using Statistical Parametric Mapping software (SPM2, http://www.fil.ion.ac.uk). After realignment, the scans were normalized to the Montreal Neurological Institute (MNI) template and smoothed spatially using a Gaussian kernel of 9 × 9 × 9 mm. Hemodynamic fluctuations related to the stochastic events were estimated subsequently with an appropriate design matrix for an event-related evaluation using the hemodynamic response function (hrf) as reference. The resulting set of voxel values constituted a statistical parametric map. We first computed the contrasts novel words versus baseline and repeated words versus baseline on the single subject level. After computing contrasts for single subjects, group data were analyzed using a random effects model (Friston et al. 1999) in order to take inter-individual variation into account [threshold P < 0.01, corrected minimum cluster size of ten adjacent voxels].
For the following subtractive contrasts (novel vs. repeated, repeated vs. novel words) paired t tests were employed. To increase sensitivity of differential contrasts between similar networks, the threshold was set to P < 0.005, uncorrected (minimum five adjacent voxels). To reduce false positive results, a subsequent small volume correction (P < 0.05) was applied to the results of the paired t tests using cubes of 20 × 20 × 20 mm size, centered at coordinates of MNI space which were chosen according to semantic activation clusters reported in a recent metaanalysis (Vigneau et al. 2006). The regions of interest were situated along the ventral pathway within left inferior frontal cortex (pars opercularis, triangularis and orbitalis) and temporal lobe (superior, middle and inferior frontal lobe). The results section only reports activation clusters showing significant activations in the small volume correction.
Summary of fMRI activations for semantic decisions versus baseline for all stimuli (P < 0.01, false discovery rate, minimum ten adjacent voxels) and the comparison novel versus repeated word stimuli (paired t test, Small Volume Correction, P < 0.05, false discovery rate, minimum five adjacent voxels)
Semantic decision repeated and novel words
Left inferior frontal gyrus
Left middle frontal gyrus
Left middle frontal gyrus
Left superior frontal gyrus
Left superior frontal gyrus
Right inferior frontal gyrus
Right middle frontal gyrus
Right superior frontal gyrus
Left anterior superior temporal gyrus
Left parahippocampal gyrus
Left supramarginal gyrus
Right parahippocampal gyrus
Right supramarginal gyrus
Left inferior occipital gyrus
Right inferior occipital gyrus
Right inferior occipital gyrus
Semantic decision novel versus repeated words
Left middle frontal gyrus
Left inferior frontal gyrus
Pars triangularis of the inferior frontal gyrus (BA45)
Pars orbitalis of the inferior frontal gyrus (BA47)
Middle frontal gyrus within BA9.
The present study provides psychophysical evidence for a robust repetition priming effect within the conceptual network even if the retention interval between first and second stimulus encounter is extended to 3 days. The functional imaging data showed a reduction of task-related functional imaging activations within the left inferior and middle frontal gyri for the comparison of repeated versus novel word stimuli.
The bilateral temporo-frontal network for abstract/concrete decisions revealed by the fast event-related fMRI design used in the present study is consistent with prior reports (McDermott et al. 2003; Binder et al. 2003; Damasio et al. 2004; Binder et al. 2005), which mainly assigned the temporal and inferior frontal activations to conceptual processing and lexical retrieval and the middle and superior frontal activations to more abstract semantic representations as well as task monitoring and response selection (Badre and Wagner 2004). The right hemispheric frontal and bilateral temporal activations showed no differential response to stimulus repetition at the chosen threshold level, indicating a great overlap of neuronal networks subserving early word processing and task monitoring for known and unknown words.
The regions which exhibited a repetition suppression in the present study were situated within the left frontal lobe. We found three distinguishable clusters of fMRI activation showing a greater activation for novel versus repeated words, two in the LIPC in the orbital part (BA47) and in the pars triangularis of the inferior frontal gyrus (BA45), the third more dorsal in the caudal part of the middle frontal gyrus (BA9), all of which have been related to semantic priming in previous studies investigating implicit memory processes within one experimental session (Demb et al. 1995; Wagner et al. 2000; Donaldson et al. 2001; Rissman et al. 2003; Copland et al. 2003). The peak coordinates of the clusters showing a repetition suppression in the present study are situated within areas reported in three previous fMRI priming studies which also used abstract/concrete decision tasks yet with a retention interval of seconds or minutes (Demb et al. 1995; Wagner et al. 2000; Donaldson et al. 2001). Thus comparison of previous reports with the results of the present study indicates that the repetition suppression found in the present study is very similar to the effect found in previous semantic priming studies using short lags. This supports the notion that short-term and long-term priming in a semantic task have similar mechanisms on the neural level. Furthermore, the present study showed that long-lasting implicit memory effects cannot only be found in the occipital, sensory network but also in the frontal network, where stimuli from different modalities converge, involving many different cognitive processes. Thus long-term priming seems to be a very stable phenomenon which also modulates neuronal responses in regions, which are not only involved in the task which is tested, but in many other cognitive processes which are not directly related.
As previous psychophysical (Friedrich et al. 1991; Vaidya et al. 1997; Thompson-Schill and Gabrieli 1999) and functional imaging studies (Demb et al. 1995) indicate, the priming effect in a semantic classification task as used in the present study is not mediated by repeated visual letter processing but lexical access is critical. Furthermore, priming effects in left inferior frontal cortex are stable across presentation modalities and perceptual forms (Wagner et al. 1997; Buckner et al. 2000). A recent study showed that Transcranial Magnetic Stimulation over the left inferior frontal cortex disrupts neural and behavioral markers of priming in an object classification task, whereas response in early sensory regions was unaffected (Wig et al. 2005). A priming-related effect in latero-temporal areas was not found in the present study. We speculate that the lack of priming-related modulation of temporal activity is related to the very long retention interval. It seems that that frontal regions are more involved in preserving implicit conceptual memory, whereas temporal regions play a minor role. Also the nature of the task involving dichotomic decisions which did not require complex conceptualizations might explain this result.
Recent studies on the priming effect provide evidence for a dissociation between explicit and implicit memory networks (Donaldson et al. 2001; Schott et al. 2005; Schott et al. 2006), suggesting separate mechanisms underlying implicit and explicit memory. The functional relevance of cortical regions exhibiting repetition suppression for the behavioral priming effect has recently been shown in functional imaging (Blasi et al. 2002; Maccotta and Buckner 2004) and transcranial magnetic stimulation studies (Wig et al. 2005). However, the concept of effectivity increases by a neuronal network tuning mechanism as physiological correlate of the priming effect has been recently challenged (Dobbins et al. 2004; Schnyer et al. 2006). These studies showed that the priming effect in an object size decision task is reversed, if the task is changed. Thus the priming effect may not be related to a tuning of the cortical networks subserving stimulus representation but to learning of an association between a stimulus and a required response. Although it cannot be entirely ruled out that response learning was involved in the behavioral priming effect reported here, we consider it unlikely that it played a major role, as each stimulus was only presented once per session, and we although observed comparable robust effects for comparison of novel with known words and of identical stimuli at first versus second encounter. In contrast, the response learning effects described by the two studies of Dobbins et al. and Schnyer et al. which used a retention interval of several minutes, were relatively weak if the target had been encountered only once before. Thus the robustness of the priming effect over several days might suggest a further underlying mechanism related to the stimulus representation itself.
In conclusion, the present study showed that repetition suppression within the left inferior frontal network related to semantic classifications is detectable even over the long period of 3 days. Accordingly, a behavioral priming effect was shown over the same time span. This result extends previous results on the longevity of repetition suppression, providing evidence for representational changes in the network involved in semantic decisions on previously encountered stimuli over long periods of time.
This research project was supported by the IZKF “BIOMAT.”—“Interdisciplinary Center for Clinical Research” of the Medical Faculty of the RWTH Aachen University (Grant N55) and START-program of the Medical Faculty of the RWTH Aachen (Grant No. START 26/05). I.G.M. is supported by the Deutsche Forschungsgemeinschaft (Grant No. ME 2104/3–1).