The authors have declared that no competing interests exist.
Conceived and designed the experiments: SG RJH AW MJ PZ CD. Performed the experiments: SG AW. Analyzed the data: SG RJH CD. Wrote the paper: SG RJH CD PZ. Designed the software used in EEG analysis: MJ.
Numerous studies have reported neurophysiological effects of semantic priming in electroencephalography (EEG) and in functional magnetic resonance imaging (fMRI). Because of differing methodological constraints, the comparability of the observed effects remains unclear. To directly compare EEG and fMRI effects and neural sources of semantic priming, we conducted a semantic word-picture priming experiment while measuring EEG and fMRI simultaneously. The visually presented primes were pseudowords, words unrelated to the target, semantically related words and the identical names of the target. Distributed source analysis of the event-related potentials (ERPs) successfully revealed a large effect of semantic prime-target relatedness (the N400 effect), which was driven by activations in a left-temporal source region. However, no significantly differing activations between priming conditions were found in the fMRI data. Our results support the notion that, for joint interpretations of existing EEG and fMRI studies of semantic priming, we need to fully appreciate the respective methodological limitations. Second, they show that simultaneous EEG-fMRI, including ERP source localization, is a feasible and promising methodological advancement for the investigation of higher-cognitive processes. Third, they substantiate the finding that, compared to fMRI, ERPs are often more sensitive to subtle cognitive effects.
The term
The number and location of identified brain regions potentially related to semantic priming varies between studies, materials, tasks and methods, but some converging evidence exists. Most notably, studies with lesion data (e.g.,
While the methods mentioned give complementary insights into the neural basis of semantic priming, they differ substantially in terms of their experimental procedures, participant samples and, in particular, with respect to the aspects of neural processing they reveal best. Therefore, it remains challenging to put the results from these different methods into perspective, and, consequently, to obtain a complete picture of the underlying neural dynamics.
One way to overcome some of these differences is to use the identical experimental procedure and the same participants in two successive measures (as was done, for example, by Matsumoto et al.
The recently developed possibility to record EEG and fMRI data simultaneously promises to minimize the above problems. In this method, the recorded data are in response to identical experimental stimulation, recorded in the identical environment, and they are based on identical underlying brain activity of the same participant
In the present study, we measured EEG and fMRI simultaneously in the context of a semantic priming experiment. For data analysis, we decided to directly compare experimental effects and source localizations between the two neurophysiological measures. We consider such a direct comparison to be valuable in and of itself, because it offers the much-needed opportunity to contrast EEG and fMRI activations to identical brain activity. Moreover, such a comparison represents a thorough empirical starting point to link the existing EEG and fMRI literatures.
In the study reported here, we implemented a priming design with words as primes and pictures as targets. There is ample evidence for priming effects in EEG
Fifteen right-handed native speakers of German participated in the study (9 female, mean age = 26.4 years, range = 22 to 29 years). All had normal or corrected-to-normal vision. One participant's dataset had to be discarded from the EEG analysis and four from the fMRI analysis due to excessive artifacts or technical problems during data acquisition.
Participants were informed about the risks, aims and procedure of the study and gave their written consent. The study was approved by the ethical committee of the German Society for Psychology (DGPs).
Based on a previous study
Four primes (one from each condition) were selected for each picture. For three of the four priming conditions, 38 word primes each were selected from the earlier study
Example primes for the four conditions are presented at the bottom left, with English translations in parentheses. Participants were instructed to press the button only after the disappearance of the target picture, i.e., during presentation of the following blank screen.
Word and picture stimuli were presented centrally on a white background using Presentation 12.2 experimental software (Neurobehavioral Systems, San Francisco, California) and an MR suitable projection system (Sharp XG-PC10XE). The eye-to-screen distance was 105 cm and all stimuli had a maximum viewing angle of 6° in height and width. Words appeared in black “Arial” font, with first letters capitalized (coherent with German orthography for nouns). Participants gave button responses with their right hand's middle and index fingers on an MR-compatible response pad.
All 38 target pictures were pseudorandomly presented four times in each of two consecutive runs (38×4×2 = 304 trials), once with each of its four prime words. The order of presentation was constrained to allow no more than three consecutive trials from the same condition and to contain an even distribution of the possible transitions between the conditions of two consecutive trials.
Simultaneously with fMRI acquisition, the EEG was recorded using an MR compatible amplifier (BrainAmp MR plus, Brain Products, Gilching, Germany). A 32-channel EEG cap (model “Easycap BrainCap-MR 3-0 32Ch) with 30 scalp electrodes located according to the international 10-10 system was employed. Of the two remaining electrodes, one was positioned on the back, left of the spinal column (between the 5th and 7th costa), to record the electrocardiogram, and one was positioned under the left eye to record the electrooculogram. To avoid uncomfortable pressure due to resting on the electrode cap and to avoid increased head movements, the participants' heads were rested on custom foam padding that included holes for electrodes and cables. During measurement, electrode impedances were kept below 15 kΩ, with FCz serving as the reference channel. The hardware clock of the EEG amplifier system was synchronized with the clock driving the MRI scanner's gradient switching system by means of a commercial device (SyncBox, Brain Products, Gilching, Germany), to guarantee the temporal stability of the EEG acquisition in relation to the switching of the gradients during the MR acquisition. The data were recorded with a pass band of 0.016–250 Hz and digitized at 5000 Hz at 16 bit with 0.5 µV resolution (dynamic range, 16.38 mV).
In a first step, EEG data were corrected for MR gradient and ballistocardiac artifacts by applying modified versions of the averaged artifact subtraction and adaptive noise cancellation algorithms proposed by Allen and colleagues
Time-locked to the picture onset, segments of 1200 ms length were then extracted from the data. Segments began 200 ms before the picture onset and lasted until 1000 ms after picture onset. Baseline correction was performed using the 200 ms interval preceding the picture.
To further reduce MR-related and common EEG artifacts, we applied extended infomax ICA
ERPs were then calculated for each experimental condition of each participant's data set. The number of trials surviving artifact rejection and correction (
Magnetic resonance imaging was performed at 3 Tesla on a Philips Medical system (Best, the Netherlands) equipped with a standard birdcage head coil, at the Institute for Clinical Radiology, University of Münster, Germany. An extended series of echo-planar-images (EPI) were gathered, including whole-head measurements before and after the experimental task. Thirty-six axial slices were obtained aligned to the anterior and posterior commissure (thickness 3.6 mm), using a multi-slice EPI sequence with an echo time of 38 ms, a flip angle of 90° and a repetition time of 3000 ms. The fMRI pixel matrix acquired was 64×64, with a field of view of 230×230 mm2, resulting in a voxel resolution of 3.6×3.6×3.6 mm3. The same parameters were also used to cover the whole brain with 43 slices. These whole-head EPIs were later used to optimize the spatial normalization of the functional echo planar images.
The basic preprocessing of the images and fMRI-statistics were done using SPM5 (
Because we were interested in the generators of neural activity itself, statistical analyses of ERP data will be presented in source space activity estimates only. Notwithstanding, as shown in
ERPs from eleven standard electrode positions are presented, plotted by priming condition in an interval stretching from 200
To retain comparability with previous localization studies
To accurately assess the sources and time interval(s) relevant to the semantic priming effect whilst avoiding circular analysis, the data were analyzed in a two-level procedure, in accordance with the independent split-data approach
For each subject, a fixed-effects analysis was applied, using a canonical hemodynamic response function (HRF) to model the four priming conditions (identical name, related word, unrelated word, pseudoword). The events for each of the four conditions were modeled by specifying the onset of the target stimulus in time and considering the events instantaneous (SPM parameter 0). In addition, movement-related parameters derived from the spatial realignment were included as covariates to further control for task-irrelevant artifacts (movements did not exceed those typically observed in fMRI-only measurements). Contrast maps of the four experimental conditions of each subject were entered into a random-effects group analysis based on the General Linear Model as implemented in SPM5. A conjoined
Participants were instructed to give responses (man or nature made) only after the target picture had disappeared. Therefore, reaction time data were not analyzed. The error rate was low and did not differ significantly across conditions,
The Global Power of the estimated neural activities across all test sources is shown in panel A of
a) Global Power of estimated EEG source activations across all test dipoles. b) Global Power of
To select the interval for second-order analysis, we opted for a data-driven approach: a pointwise one-way repeated measurement analysis of variance (ANOVA) with the 4-level factor
Data from the second half of participants (
Statistical parametric maps from the random effects group analysis of fMRI data are depicted in
a) Coronal, sagittal and axial view for the whole-head contrast of the effect of stimulus presentation against the implicit baseline. b) Whole-head contrasts of the individual priming conditions against the implicit baseline in coronal and axial views (visualizations correspond to
Cluster size | MNI coordinates | ||||||
H | Anatomical region | BA | (voxels) | x | Y | z | |
L | post-central | 4 | 18.91 | 497 | −60 | −8 | 46 |
L | superior temporal gyrus | 22 | 29.89 | 1907 | −58 | −22 | 9 |
L | insula | 48 | 38.35 | 1613 | −44 | −2 | 0 |
L | putamen | 48 | 22.67 | 1103 | −30 | 4 | 8 |
L | cerebellum | 18/19 | 20.49 | 511 | −20 | −86 | −20 |
B | (pre-) supplementary motor area | 6 | 23.10 | 616 | −6 | 0 | 60 |
B | dorsal anterior cingulate | 24/32 | 23.22 | 615 | 10 | 16 | 36 |
R | insula | 48 | 24.39 | 314 | 42 | −2 | 10 |
R | putamen | 48 | 19.62 | 311 | 22 | 14 | 4 |
R | cerebellum | 18/19 | 31.32 | 2087 | 18 | −58 | −22 |
When comparing the individual conditions using fMRI, it seemed that activation patterns were virtually identical to the overall effect. Indeed, statistically comparing the pseudoword against the three word priming conditions (pseudoword – unrelated, related, and identical, respectively) by means of
Two steps were taken to further address whether priming-related differences corresponding to those seen with the EEG might simply be less pronounced yet still present in the fMRI data. First, whole-head fMRI
Furthermore, to investigate the relationship between the fMRI and ERP activations, we calculated correlations of the experimental effect in the identified ERP sources with those from selected fMRI clusters, similar to the approach taken by Matsumoto and colleagues
In the present study, we simultaneously recorded EEG and fMRI during a semantic word-picture priming experiment to compare experimental effects and neural sources from these two measures. We performed distributed source analysis on the ERP data using L2 minimum norm estimation, and compared the results to fMRI data that were analyzed using statistical parametric maps derived from a general linear model based on the canonical HRF.
As predicted, EEG source activity during the N400 interval was strongly modulated by the semantic similarity or distance between prime words and target pictures: The larger the semantic distance, the larger the N400 amplitude. This gradation of N400 amplitudes along semantic distance corresponds to previous results from EEG and MEG studies, with pictures and with words as targets
Distributed source analysis of the ERP data based on the L2 minimum norm procedure revealed that the sources showing the strongest priming effect were situated in a confined region of the left temporal lobe. Second-level analysis within this left-temporal and a contralateral cluster showed that the differentiation of N400 source amplitudes among priming conditions followed the predicted order in a linear fashion. The observed ERP localization corresponds with the general pattern from previous MEG and EEG studies on semantic priming that were performed outside the MR scanner
In the fMRI data, whole-brain analysis revealed widespread
The current study was designed to replicate the MEG effect from an earlier word-learning study
With regard to the fMRI literature, there are several studies that used word-picture presentations in picture-naming paradigms
A further difference to previous fMRI studies concerns stimulus repetition. Our experimental design deliberately included repetitions of primes (2 presentations) and targets (8 presentations), while most fMRI studies were designed to avoid stimulus repetitions
A third difference concerns response instruction. Most of the existing fMRI studies on semantic priming
Taken together, design differences between the current study and existing studies on semantic priming make it difficult to single out one factor that may have led to the observed null result in fMRI. However, we think that these factors should be investigated more closely, such that we obtain a more generalized understanding of semantic priming and its neural correlates.
Independent of the differences between the current fMRI results and those from the literature, it is worth taking a closer look at the discrepancy between the fMRI and ERP effects within the current study. In their review, van Petten and Luka
To address phase resetting in the context of the N400, Mormann et al.
Beyond this theoretical explanation, Vartiainen and colleagues
In conclusion, in this semantic word-picture priming experiment we obtained robust ERP effects in the N400 window, and the EEG source analysis clearly replicates findings of a predominantly left-lateralized temporal origin for these effects. We found little evidence that fMRI activations similarly distinguish between priming conditions. The fact that the EEG effect was considerable, despite of the simultaneous fMRI recording, underlines the usefulness of a combined registration for the investigation of higher cognitive processes, and their potentially different signatures in different brain measures.
We thank Thore Apitz, Verena Bremer and Harald Kugel for their assistance in the recording and analyzing of the data.