Atypical language organization in temporal lobe epilepsy revealed by a passive semantic paradigm

Temporal lobe epilepsy (TLE), the most common cause of intractable epilepsy in adults, can be successfully cured by surgery [1]. Partial removal of the anterior temporal lobe (ATL) remains the most common form of surgical treatment and is effective in 60-80% of patients [1‐3]. A complication of surgery observed in 30-50% of patients after left ATL resection is a decline in language or verbal memory function [4‐7]. The magnitude of decline is related to the degree of language lateralization to the left hemisphere and can be predicted using preoperative functional magnetic resonance imaging fMRI; [6‐8].

A decrease of regional cerebral metabolism unilateral to the epileptogenic focus has been observed in patients with intractable TLE, usually in temporal and occasionally prefrontal regions [9]. Similarly, the reorganization of language may also affect different productive (located in frontal areas) and perceptive (located in temporal regions) language functions mainly in temporal rather than frontal areas [10, 11]. Atypical functional lateralization for refractory epileptic patients during fMRI language tasks has been widely reported with either bilateral or right lateralized patterns of activation [12]. This is predominant in patients suffering from left mesial temporal lobe epilepsy LMTLE; [6, 11, 13‐15].

Although the tasks employed by epilepsy institutes vary greatly, most fMRI paradigms used to determine the language dominance in epileptic populations include language production paradigms such as word or verb generation [11, 13, 15‐18], object naming [14], or sentence repetition [11]. More complex paradigms involving semantic decisions have also been used [6, 7, 12, 18, 19]. All these paradigms require the patient to perform an active action (e.g., naming a word, making a semantic decision) which critically depends on his/her capacity to collaborate during the execution of the task. Paradigms that are too complex might result in patient underperformance, yielding poor activation patterns.

Surprisingly, only a few studies of epilepsy sufferers have used pure passive tasks such as listening to standard speech [11, 20, 21]. Passive listening paradigms reliably activate the receptive language cortex [22] making it possible to determine hemispheric dominance and identify the areas involved in language processing of TLE patients, which is of the utmost importance in pre-surgical language mapping [10, 11]. These tasks commonly bring about activations in Wernicke's area –mainly the superior temporal gyrus (STG) and the middle temporal gyrus (MTG), frequently extending into the angular gyrus- but also in the primary auditory area and less often in frontal regions [11, 20, 23, 24]. A study comprising healthy children [24] showed that both passive listening and active-response story processing yielded activations in the primary auditory cortex and the STG bilaterally and also in the left inferior frontal gyrus (IFG). It has also been shown in TLE patients with atypical language patterns that plastic cortical reorganization can affect frontal and temporal language areas differently with greater reorganization of language circuits in the temporal rather than the frontal lobe [11, 20]. This suggests that productive and receptive functions can be affected by the pathological process in different ways. In contrast with productive tasks, passive tasks are easy to perform, less dependent on patient collaboration and allow the assessment of how fronto-temporal networks contribute to other aspects of language processing. In addition, the pattern of activation of some receptive language paradigms can be as lateralized as that of verbal fluency or semantic decision tasks [11, 20, 22‐25]. However, it is also well-known that using a baseline condition such as fixation or rest can lead to more bilateral activations [26].

The main purpose of this study is to investigate the reliability of a passive, non-collaboration dependent, semantic fMRI language task in evaluating language lateralization patterns in a group of selected left and right mesial temporal lobe epilepsy (RMTLE) patients. These results were compared to those of a well-matched healthy group. Based on previous studies, we expected to observe a clear increase of right-hemisphere lateralization associated to earlier age at epilepsy onset in LMTLE patients [12, 27].

In line with previous intracarotid amobarbital procedure data and consistent with our findings, several fMRI studies have suggested that atypical dominance patterns are more frequent in epileptic patients [12, 17, 19, 40] and may predominate in patients with epilepsy of left hemispheric origin [13, 19, 41]. A previous study [41] showed that 24% of LMTLE patients with HS had atypical language dominance and more importantly that this atypical language representation was associated with both a higher frequency of interictal discharges and with sensory auras representing seizure propagation to lateral temporal structures. It has been suggested that propagation of seizures from the epileptic focus to different brain regions triggers cortical plasticity not only around the focus but also in remote areas either ipsilateral or contralateral to the epileptic focus [13, 14, 17, 19, 40]. It would therefore appear that the language network is particularly vulnerable to chronic epileptic activity. Moreover, an inter-hemispheric dissociation of frontal and temporal language regions in patients with focal epilepsy has also been described [10, 11, 20, 42‐44]. In a previous study, 29 (20.1%) out of 144 patients with medically intractable complex-partial seizures showed bilateral language representation after intracarotid amobarbital test assessment [10] and more importantly, 4 (2.8%) of these patients -2 of them with TLE- had strong evidence of an interhemispheric dissociation of expressive and receptive language functions. These findings have been posteriorly confirmed by several functional neuroimaging studies [11, 20, 42‐44], which all support the hypothesis that reorganization can affect frontal and temporal language areas differently in patients with atypical language patterns.

Paradigms used for clinical language mapping vary greatly in different centres and no accepted standardized technique exists. Most fMRI language dominance studies in epileptic patients have used active language tasks [13‐18]. However, imaging studies have confirmed that the laterality indexes yielded by receptive language paradigms can be as lateralized as those of active tasks [11, 20, 22‐25]. In this way, and similarly to other passive listening semantic paradigms used on TLE patients [11, 20, 21], we show how passive contrasts such as listening to SENs or PSENs versus baseline are able to unravel and clarify atypical language patterns in LMTLE patients. Mean LI was significantly more right lateralized for LMTLE patients than controls when listening to SENs and again more right lateralized for LMTLE than for both controls and RMTLE patients when listening to PSENs (see Figure 1 and Table 1). While LI was mainly bilateral for the controls (79% of subjects for both SEN and PSEN conditions) and for RMTLE patients (81.8% of patients, for both SEN vs. rest and PSEN vs. rest conditions), 6 out of 12 patients of the LMTLE group showed a right lateralized activation for both contrasts (50% of patients).

It is important to note that this passive listening task is very easy and rapid to perform (only 6 minutes of fMRI acquisition) and that both task-rest contrasts produced robust activations of the temporal lobe for every one of the 42 participants studied. Interestingly, the SEN vs. PSEN contrast failed to elicit robust responses in 42% of LMTLE patients and controls and in 36% of RMTLE patients. These results are consistent with other passive paradigms using direct speech contrast between two active conditions, which also failed to show significant activations in several subjects. In a previous study [11], a direct contrast between two active conditions (Listening Stories vs. Listening Reversed Stories) produced no significant activation in 22% of RMTLE and LMTLE patients and 12% of healthy subjects. Interestingly, in another study [35], activation in 100% of subjects failed to reach significance in the direct contrast between two active conditions (listening to Words, Pseudowords and Reversed Speech).

In accordance with the present results, there is extensive functional imaging data suggesting that language comprehension is subserved by a large cortical network involving both temporal cortices [45‐52]. Many neuroimaging studies reveal weak neural activity in the right hemisphere in the anatomically equivalent areas to those of the left hemisphere which show a strong signal during language tasks [48, 51]. Additionally, some patients with right brain damage have deficits in comprehending natural language [48]. All these data support the idea that although the left hemisphere plays a crucial role in language processing, the right hemisphere also contributes to language comprehension [45, 48, 50, 51, 60]. Hence, the bilateral pattern of temporal lobe activation in controls and MTLE patients using this passive semantic paradigm agrees with previous functional imaging data showing that the STG and MTG are activated bilaterally by both speech and complex non-speech sounds [20, 32, 35, 45, 53‐61]. Even when the auditory stimuli is novel and no meaning extraction is possible (as in the PSEN condition) both the MTG and STG show enhanced activations [62, 63]. In fact, some studies have shown that activation at the level of the auditory cortex is neither modulated by the semantic content ("meaningfulness") of stimuli nor by the type of cognitive task performed [64]. Moreover, in a previous study [35] Words, Pseudowords and Reversed Speech versus baseline bilaterally activated both temporal lobes, yielding less than 2 ml of volumetric difference in activation between hemispheres (the only contrast that showed a significant difference in activation in temporal lobes was Words vs. rest with an LI of 0.11, therefore showing a bilateral distribution). It is also well known that that using a baseline condition such as fixation or rest (as in the SEN vs. rest and PSEN vs. rest contrast) yields bilateral functional patterns and LI values around zero [26]. There is also experimental and clinical evidence showing bilateral activation in conceptual processing [65].

Furthermore, a significant correlation between LI and age at epilepsy onset was found for the PSEN condition versus baseline (LIs for SEN vs. rest contrast also yielded a trend). There is a gradual decline in the potential for plasticity with age [12, 27], but no particular age after which plasticity becomes absolutely impossible [66]. Visual inspection of patients with LMTLE and early onset clearly showed a right lateralized pattern (Figure 3A), while subjects suffering from RMTLE or LMTLE with late epilepsy debut showed a more bilateral pattern (Figure 4B and 4C, respectively). It is important to note that a correlation between age at epilepsy onset and LI has been previously described using a semantic decision task [12]. This gives additional support to the usefulness of a simple passive language task like the one described in this study.

Previous studies using language fMRI with productive and receptive tasks have shown how in TLE patients with atypical language patterns, reorganization affected frontal and temporal language areas differently, with temporal LIs being more right lateralized than frontal ones [10, 11, 20, 42‐44]. Moreover, data from lesion and neuroimaging studies suggests that comprehension or semantic activation -involving retrieval and selection of semantic information- depends on a wide cortical network, larger than Wernicke's or even Broca’s area (which is required particularly for the correct processing of complex morphosyntactic structures) [67]. Furthermore, there are several patient studies supporting the involvement of the anterior superior temporal lobe for sentence-level comprehension [45, 65, 67, 68]. Semantic knowledge stores are diffusely located in ventral temporal (MTG, ITG, fusiform gyrus, temporal pole) and inferior parietal (angular gyrus) cortices, in addition to the classic posterior STG and planum temporale [23, 45, 48, 50, 69, 70]. Therefore, as both the ventral and anterior temporal lobes are related to receptive language and TLE surgery (ATL) commonly involves the resection of these regions of the temporal lobe, we hypothesize that the passive semantic paradigm presented here -which mainly activates temporal lobe structures- is more precise in predicting semantic language defects, such as naming retrieval or verbal memory decline, in TLE patients who undergo surgery [4, 23, 70‐76]. In addition, this paradigm may also provide important information about the extent of functional reserves in the contralateral hemisphere [20] and might be helpful in tailoring the resection of temporal lesions with the aim of preserving eloquent areas of the brain. We cannot omit however that a limitation of this study is the lack of comparison to other laterality measurements yielded by more classical non-passive language paradigms. Also, a larger sample of patients might have allowed us to compute a more detailed analysis of the LIs within the temporal lobes by dividing the MTL into different sections (i.e., anterior/posterior).

In summary, the simple, rapid, non-collaboration dependent, passive task described in this study produces a robust activation in the temporal lobe in both TLE patients and controls and illustrates a pattern of atypical language organization specifically in LMTLE patients.