Assessment of verbal memory by fMRI: Lateralization and functional neuroanatomy

2.1. Subjects 

Ten healthy volunteers (5 men), aged 23–28 years, participated in the study. Written informed consent was obtained prior to participation according to the declaration of Helsinki. All participants were native German speakers, right-handed according to the Edinburgh handedness inventory [16] and had completed the equivalent of a high school degree (“Abitur”). None of the subjects had a history of neurological or psychiatric illnesses, brain pathology or abnormal brain morphology on T1-weighted MR images.

2.2. Paradigm 

Verbal memory functions were investigated by comparing activation differences between the encoding of known and unknown verbal stimuli. As stimulus material, common German words were selected from the database of the CELEX Corpus (http://www.ru.nl/celex). All words had a medium word frequency, 2 syllables, and 6–7 letters. Half of the words were verbs, the other half nouns.

Stimuli were presented visually in capital letters. Subjects were instructed to remember all words for a later recognition test. They were not informed about the existence of different conditions or presentation frequencies.

The imaging session started with a 5-min anatomical MR-image. During this time, the subjects repeatedly viewed 6 words that were later used as “old” stimuli. In the actual fMRI session, subjects were presented with 60 “old” (each old stimulus was thus shown ten times) and 60 “new” stimuli (each new stimulus was shown only once) in alternating blocks of variable length. Stimuli were presented for 2s, with an interstimulus interval of 2s. The epoch length varied between 1 and 8 stimuli, with an average block length of 5 stimuli. To ensure a high level of attention, the participants were instructed to decide whether the stimulus shown was a verb or a noun by pressing one of two buttons of a response box using their right hand.

After fMRI scanning, subjects performed a recognition memory test, in which the same 60 new stimuli were randomly presented along with 60 other distractor words (“recognition task”). Participants were instructed to indicate whether or not each word had been presented during the fMRI session.

2.3. MRI data acquisition and analysis 

All MRI data were acquired on a Philips 3T whole body scanner equipped with a standard circular polarized head coil. Functional images were acquired using a T2* weighted spin echo EPI sequence (TE=35ms, TR=3.5s, flip angle 90°, slice thickness 3.6mm without gap, matrix 64×64, FOV 230mm, in-plane resolution 3.6mm×3.6mm, 36 axial slices orientated parallel to the AC-PC line covering the whole brain).

SPM2 (http://www.fil.ion.ucl.ac.uk/spm) standard routines and templates were used for analysis of fMRI data. The functional images were temporally and spatially realigned, normalized (resulting voxel size 2mm×2mm×2mm), smoothed (6mm isotropic Gaussian filter for group analysis, 4mm for individual analysis) and high-pass filtered (cut off period 120s). Statistical analysis was performed in a two-level, mixed-effects procedure. In the first level, the BOLD responses for the encoding of new and old items, respectively, were modeled by a hemodynamic response function. Parameter estimate (β-) and t-statistic images were calculated for each subject. In the second level, the individual β-contrast images were used to determine activations at the group level. The anatomical localization of brain activity was identified using the software packages Automated Anatomical Labelling (AAL) [17] and Münster T2T converter [18].

The lateralization of MTL-activity was assessed on the one hand by the thresholded activation pattern, on the other hand by calculation of lateralization indices.

Thresholded activation pattern: MTL-activity was assessed at different statistical thresholds by sequentially liberalizing the p-value (p=0.001, 0.01, 0.05, uncorrected for multiple comparisons). The MTL was defined as the hippocampus, the parahippocampus, and the part of the fusiform gyrus along the full length of the hippocampus, but excluding the amygdala [11].

Calculation of a lateralization index: Hemispheric dominance is typically described by a lateralization index (LI). Several approaches have been established (for a discussion, see [19], [20]). We calculated the LI by the formula

3.1. Behavioural data 

During the fMRI experiment, more than 98% of the stimuli were judged correctly as a verb or a noun. Reaction times were shorter for “old” stimuli (683ms) than for “new” stimuli (771ms). During the recognition test, 71.1% of words were judged correctly. Subjects were significantly better to categorize an unknown stimulus as new (79.9% correct responses) than a known stimulus as familiar (62.3% correct responses). This is backed up by the reaction times (906ms vs. 938ms).

3.2. Brain activation: group results 

For the contrast NEW>OLD, the group analysis revealed mostly left-lateralized brain activity, with main activation centres located in the prefrontal cortex, the supplementary motor area, the inferior parietal cortex, the MTL and the cerebellum. For the contrast OLD>NEW, no brain regions were active at p=0.001 uncorrected.

In the MTL, three activation clusters, two located in the anterior and one in the posterior MTL, were distinguishable (Fig. 1). The activity was unambiguously left-lateralized, as no brain activation was visible in the right MTL.

View Large Image Download to PowerPoint

Fig. 1. Three clusters were active in the MTL during word encoding (random effects analysis, contrast NEW>OLD, p=0.001 uncorrected). The activation is left-lateralized, no brain activation could be found in the right MTL (although at more liberal thresholds, a small activation cluster homologous to activation 3 became visible, p>0.01 uncorrected).

3.3. Brain activation: individual results 

At p=0.001, three subjects showed brain activation of the MTL (subjects 1, 4, 7). Different from the group results, the activation consisted for each subject only of one cluster. This cluster was left-lateralized in all cases; no activation was found in the right MTL.

The anatomical localization of MTL-activation was highly variable between subjects. The individual clusters were between 7 and 33mm apart (for individual MNI coordinates see Table 1). Furthermore, it was not possible to assign any of the individual activation clusters unambiguously to one of the three cluster coordinates from the group analysis. The distances between the individual activation cluster and one of the three group activation clusters differed between 9 and 40mm.

Table 1. MNI coordinates of MTL-activity in subject 1, 4, and 7 and its distance (in mm) to the activation clusters 1–3 obtained by the group analysis

	Subject	MNI coordinates			Distance (in mm) to group activation cluster no.
		X	Y	Z	#1	#2	#3
	1	−14	4	−32	40	25	9
	4	−32	−24	−18	10	18	34
	7	−32	−20	−22	15	13	30

When the statistical threshold was sequentially liberalized to p=0.01 and p=0.05, left-lateralized MTL-activation was found in eight subjects, right-lateralized activation in one subject. Again, the activation consisted for each subject only of one cluster. A clear assignment of MTL-activations to any of the group activation clusters was not possible.

3.4. LI calculation 

LIs were calculated in ROIs centred on the 3 group activation maxima. Overall, the results were inconsistent. Many subjects were classified as left-hemisphere dominant by one method of calculation, right-hemisphere dominant by another (data not shown). Thus, the formal LIs did not allow for firm conclusion about hemispheric dominance.

4.1. The paradigm 

We selected the paradigm to suit the following demands: First, it should be sensitive to encoding of new information, and second, it should be applicable to patients in routine clinical settings. We focused on the encoding of verbal stimulus material, since impairments in verbal memory are most relevant for everyday life. We used stimuli that were either “new”, that is, only shown once, or “old”, that is, shown several times. Assuming that the encoding of known stimuli poses less demand on the neural network underlying the memory system, the contrast NEW>OLD should reveal which brain regions are involved in the encoding of information (“novelty encoding”; for a discussion of other memory paradigm principles see [13]).

We employed a “semi-random” block design that combined the accuracy and flexibility of an event-related design with the power of a block design [11], [14], [21]. The epoch length varied between 1 and 8 stimuli, to avoid that subjects anticipated the change between blocks of “new” and “old” stimuli. To ensure subjects’ participation and attention, a verb-noun discrimination task accompanied the encoding phase. Due to its simplicity, this task is also applicable to patients with altered cognitive abilities.

4.2. Functional imaging results 

At the group level and consistent with previous reports on verbal memory encoding, we found brain activation in the prefrontal cortex, the supplementary motor area and the parietal cortex [4], [7]. The MTL-activation was strongly left-lateralized and separated into three clusters, located in the anterior as well as in the posterior MTL.

The terms “medial temporal lobe”, “hippocampus” and “hippocampal formation” are in many studies used interchangeably. This obscures the fact that the MTL is anatomically and functionally highly differentiated. Its anatomical subregions (hippocampus, entorhinal and parahippocampal cortex) contribute differentially to memory processes; their specific functions are currently under debate [22], [23]. The separation of MTL-activity in the present study thus indicates the involvement of different mnemonic operations during the encoding of verbal material.

At the individual level, the lateralization of memory-related MTL-activity could be determined in 9 of 10 subjects. The thresholded activation patterns indicated left-lateralized MTL-activity in eight subjects, right-lateralized activity in one subject. In contrast, the location of brain activity in the MTL was highly variable. Individual localizations differed not only between subjects, but also from the activation clusters obtained by the group analysis. A clear assignment of MTL-activations to any of the group activation clusters was not possible.

Furthermore, in single subjects only one activation cluster was detected in the MTL, different from the random effects group analysis. Since the purpose of a random effects analysis is to find the areas that are activated in much the same way in all subjects, it cannot be excluded that, due to the lower signal-to-noise ration in individual subjects, not all relevant activity is detected.

We also assessed the lateralization of MTL-activity by a formal calculation of LIs. The LIs were not only highly variable between subjects, but also within subjects when different activity measures were applied. This can be explained by the individual differences in functional neuroanatomy. Even if the individual peaks of MTL-activation clusters are only slightly dispersed across different locations, relatively small ROIs (as in this study) do not “catch” all relevant activity in many subjects.

4.3. MRI in the clinical assessment of memory functions 

The determination of memory capacity and memory reserve in the MTL is a critical question in the pre-operative assessment of patients with MTL epilepsy who are candidates for surgical resection of the epileptogenic foci. Before investigating patients however, it must be first assessed to which extent the verbal memory paradigm can be used to determine the lateralization and the functional neuroanatomy of MTL-activity. Therefore, this study took a “step backwards” by first examining healthy subjects without known MTL pathology.

While many studies report robust memory-related MTL-activation averaged across groups, e.g. [5], [7], [14], only few report stable results in individual subjects, e.g. [4], [11], [15]. Group results however may be sufficient for neuroscientific questions (e.g. whether hemispheric lateralization for memory encoding generally differs between men and women [24]), but not for clinical assessments (e.g. for the decision whether or not to undertake resective surgery of the MTL). For clinical assessments and surgical planning, memory lateralization and functional neuroanatomy must be determinable for single subjects.

We showed that the lateralization of memory-related MTL-activity could be assessed in most subjects. Before being applied in a clinical setting, the paradigm must now be tested in clinical populations to examine whether this finding can be replicated, for instance, also in patients with epilepsy.

The existence of three distinct activation clusters in the group analysis shows an involvement of different MTL sub-regions in verbal encoding. In contrast, only one activation cluster is detected in individual subjects (probably due to the lower signal-to-noise ratio), making it likely that important activation is “missed” on the single subject level. Epilepsy surgery has to remove the seizure focus completely, but without causing damage to parts of the MTL that are essential for verbal encoding. The present results therefore show that, if the paradigm is applied in epilepsy patients, individual results must be treated with care if it should be used to support decisions as to how far the resection of one MTL can be extended.

In the present study, we were interested in the comparison of group and single subject results. Therefore, we worked with normalized data. It should be kept in mind however that the data analyses procedures might have to be adapted in patients. Decreased normalization accuracy in patients may be a potentially important confounder of template-based fMRI analyses in the hippocampus and MTL [25]. Therefore the analysis of fMRI data without using normalization procedures and data smoothing might be advantageous [26]. Nevertheless, our approach constitutes an important complementary approach, since it yields information about the underlying activation in the MTL in the healthy population, using a group analysis approach.