Identifying cortical first and second language sites via navigated transcranial magnetic stimulation of the left hemisphere in bilinguals

Brain & Language 168 (2017) 106–116

Contents lists available at ScienceDirect

Brain & Language journal homepage: www.elsevier.com/locate/b&l

Identifying cortical first and second language sites via navigated transcranial magnetic stimulation of the left hemisphere in bilinguals Lorena Tussis a,1, Nico Sollmann a,b,1, Tobias Boeckh-Behrens c, Bernhard Meyer a, Sandro M. Krieg a,b,⇑ a

Department of Neurosurgery, Klinikum rechts der Isar, Technische Universität München, Ismaninger Str. 22, 81675 Munich, Germany TUM-Neuroimaging Center, Klinikum rechts der Isar, Technische Universität München, Munich, Germany c Section of Neuroradiology, Department of Radiology, Klinikum rechts der Isar, Technische Universität München, Ismaninger Str. 22, 81675 Munich, Germany b

a r t i c l e

i n f o

Article history: Received 13 December 2015 Revised 20 August 2016 Accepted 26 January 2017

Keywords: Bilingualism Cortical mapping Dominant hemisphere Language Transcranial magnetic stimulation Object naming

a b s t r a c t The cortical areas that code for the first (L1) and second language (L2) in bilinguals have still not been sufficiently explored. Thus, this study investigated the left-hemispheric distribution of the L1 and L2 using repetitive navigated transcranial magnetic stimulation (rTMS), in combination with an objectnaming task, in 10 healthy, right-handed volunteers. In particular, higher error rates (ERs) were observed in the L1, and there was a statistically significant difference between the ERs of L1 and L2 for no-response errors (L1 mean 11.9 ± 9.0%, L2 mean 6.5 ± 5.2%; p = 0.03). Furthermore, language-specific and shared cortical distribution patterns for the L1 and L2 were observed within the frontal, parietal, and temporal lobes with a trend towards higher occurrence of language-specific spots within posterior regions. Overall, the L1 presented a more stable pattern of language distribution compared to the L2. Ó 2017 Elsevier Inc. All rights reserved.

1. Introduction Currently, no universal definition of bilingualism is available (Blom, Kuntay, Messer, Verhagen, & Leseman, 2014). In this study, we adopted the definition given by Kohnert (2010), according to which bilinguals are ‘‘individuals who receive regular input in two. . .languages during the most dynamic period of communication development—somewhere between birth and adolescence’’ (Kohnert, 2010). A variety of studies have investigated which brain structures are required for first (L1) and second language (L2) processing in bilinguals; however, no clear results on which cortical areas are involved have been obtained yet. The majority of neu-

Abbreviations: CPS, Cortical Parcellation System; DTI FT, diffusion tensor imaging fiber tracking; DT, display time; EHI, Edinburgh Handedness Inventory; ER, error rate; fMRI, functional magnetic resonance imaging; IPI, inter-picture interval; L1, first language; L2, second language; LH, left hemisphere; MRI, magnetic resonance imaging; PTI, picture-to-trigger interval; RMT, resting motor threshold; rTMS, repetitive navigated transcranial magnetic stimulation; VAS, Visual Analogue Scale. ⇑ Corresponding author at: Department of Neurosurgery, Klinikum rechts der Isar, Technische Universität München, Ismaninger Str. 22, 81675 Munich, Germany. E-mail addresses: [email protected] (L. Tussis), [email protected] (N. Sollmann), [email protected] (T. Boeckh-Behrens), Bernhard.Meyer@ tum.de (B. Meyer), [email protected] (S.M. Krieg). 1 These authors contributed equally. http://dx.doi.org/10.1016/j.bandl.2017.01.011 0093-934X/Ó 2017 Elsevier Inc. All rights reserved.

roimaging and intraoperative stimulation research suggests that, in addition to common brain areas that code for both L1 and L2 processing, some cortical regions are specific to either L1 or L2 processing (Calabrese et al., 2001; Chee, Hon, Lee, & Soon, 2001; Chee, Tan, & Thiel, 1999; Dehaene et al., 1997; Illes et al., 1999; Lucas, McKhann, & Ojemann, 2004; Pouratian et al., 2000; Roux & Tremoulet, 2002). For example, Lucas et al. (2004), who performed intraoperative stimulation in 25 bilingual patients and 117 monolingual control patients, reported on language-specific sites in the inferior frontal area and posterior inferior parietal area (Lucas et al., 2004). Comparable to these findings, Roux and Tremoulet (2002) were also able to map language-specific sites within frontal and temporo-parietal areas intraoperatively (Roux & Tremoulet, 2002). Furthermore, a case report using intraoperative optical imaging reported on language-specific sites within the supramarginal and precentral gyrus (Pouratian et al., 2000). Regarding studies using functional magnetic resonance imaging (fMRI), Chee et al. found increased activation in opercular regions for L2 (Chee et al., 2001), whereas Dehaene et al. showed that areas coding for L1 were primarily located in the left temporal lobe with L2specific areas being highly variable within temporal and frontal areas (Dehaene et al., 1997). However, not all studies or paradigms of investigation have been able to resolve spatial separations between L1 and L2 (Chee et al., 1999; Illes et al., 1999).

L. Tussis et al. / Brain & Language 168 (2017) 106–116

Taking these findings together, it seems that the two languages in bilinguals are at least partially processed by different cortical areas. However, it is still difficult to draw definite conclusions from the research that has been conducted, as the results are derived from various studies in which the subjects differ in their age of L2 acquisition, their proficiency, and the context of the language acquisition, which are all factors that seem to impact the extent of the cortical overlap of the L1 and L2 (Blom et al., 2014). Furthermore, the tasks that the enrolled subjects had to perform in the studies varied, including phonological, word fluency, cue-word generation, working memory, and object-naming tasks. Because different tasks probably involve different brain regions (Buckner, Raichle, & Petersen, 1995), it is difficult to have a clear understanding of which functions are solved by different regions in bilinguals. Moreover, most of the studies conducted on bilingualism have used neuroimaging techniques, which, despite providing valuable insights, do not directly reveal a causal link between the regional activations observed and the functional processes involved. Thus, this study uses left-hemispheric repetitive navigated transcranial magnetic stimulation (rTMS) during an object-naming paradigm to specifically explore potential differences between task performance in L1 and L2 among healthy bilinguals within the scope of two hypotheses: (1) The L1 and L2 show different cortical distributions in the left hemisphere (LH), and (2) while the L1 shows a similar and stable pattern in all volunteers, the L2 shows a more varying pattern compared to the L1.

2. Material and methods 2.1. Study design The study was prospective and non-randomized.

2.2. Ethics approval statement This study was carried out in accordance with the recommendations of our local ethics committee (registration number: 2793/10). All of the subjects gave written informed consent in accordance with the Declaration of Helsinki.

2.3. Participants Ten healthy bilingual volunteers (7 females and 3 males, median age 23 years) took part in the study. The participants had to meet the Kohnert (2010) definition of bilingualism (Kohnert, 2010), as stated initially. As such, the participants were required to have acquired their L2 before the age of 10 years according to self-report. To check for comparable proficiency in L1 and L2 regarding the objects shown during the naming task, all volunteers underwent baseline testing without stimulation twice before rTMS-based mapping, but no other assessment of proficiency concerning bilingualism was carried out. Other inclusion criteria were age above 18 years, righthandedness according to the Edinburgh Handedness Inventory (EHI), and a written consent form. The exclusion criteria were left-handedness, less or more than two languages acquired before the age of 10 years, previous seizures, general rTMS exclusion criteria (e.g., pacemaker, cochlear implant, deep brain stimulation electrodes), and pathological findings on cranial imaging. Furthermore, volunteers that showed a difference of more than 13 correctly named objects (10% of the overall amount of presented objects) between L1 and L2 during baseline testing were excluded.

107

2.4. Magnetic resonance imaging All of the subjects underwent magnetic resonance imaging (MRI) before rTMS mapping. Scanning was conducted using a 3 T MRI scanner (Achieva 3 T, Philips Medical Systems, The Netherlands B.V.), in combination with an 8-channel phased-array head coil. A three-dimensional gradient echo sequence (TR/TE 9/4 ms, 1 mm3 isovoxel covering the whole head, 6 min and 58 s acquisition time) was acquired without intravenous contrast agent. 2.5. Language mapping The subjects’ LH was mapped twice, 14 days apart, in randomized order regarding the L1 and L2. The procedures for both mappings were the same. First, the three-dimensional MRI data of each subject were co-registered to the volunteer’s cranium, in order to provide a navigational template for rTMS mapping, which was conducted with the Nexstim eXimia NBS system (version 4.3) with a NexSpeechÒ module (Nexstim Plc., Helsinki, Finland). The system tracks the coil’s position with respect to the head using a stereotactic camera, which senses both the coil and the reflectors positioned on a strap tied to the subject’s head. The locations of the induced field and stimulation spots were displayed on the MRI data and recorded for further analysis (Ilmoniemi, Ruohonen, & Karhu, 1999; Krieg et al., 2016; Ruohonen & Karhu, 2010; Sollmann et al., 2014). Prior to language mapping, the optimal stimulation intensity for each volunteer was established by determining the resting motor threshold (RMT), which was obtained by stimulating the cortical representation of the contralateral abductor pollicis brevis muscle with decreasing intensities, following previously described protocols (Krieg et al., 2012; Picht et al., 2009, 2012). During language mapping, the subjects performed an object-naming task consisting of 131 colored photographs of everyday objects (Krieg et al., 2016; Picht et al., 2013; Sollmann et al., 2014), which appeared on a screen that was located approximately 60 cm in front of the volunteer. A video camera with a built-in microphone recorded the task performance of each individual during baseline testing and rTMS mapping (Hernandez-Pavon, Makela, Lehtinen, Lioumis, & Makela, 2014; Krieg et al., 2016; Lioumis et al., 2012; Picht et al., 2013; Sollmann et al., 2014). Recording was started immediately before the presentation of the first object on the screen, and it lasted until screening of the last object, shown during application of the final stimulation burst according to our stimulation protocol, was finished. During mapping, the inter-picture interval (IPI) was 2500 ms, the display time (DT) of each object was 700 ms, and the train of the stimuli was delivered simultaneously with picture presentation in a time-locked fashion, with a picture-to-trigger interval (PTI) of 0 ms (Krieg et al., 2014). The stimulation was performed with 100% RMT at a frequency of 5 Hz/5 pulses, with the coil oriented anterior-posteriorly, as published earlier (Krieg et al., 2016; Rosler et al., 2014; Sollmann et al., 2014; Tarapore et al., 2013). This protocol has demonstrated to be easily tolerable and efficient in terms of elicitation of naming errors during rTMS-based language mapping, and can currently be regarded as the most common approach (Krieg et al., 2016; Rosler et al., 2014; Sollmann et al., 2014; Tarapore et al., 2013). The subjects were asked to name clearly and as quickly as possible each of the 131 objects that appeared sequentially on the screen (Krieg et al., 2016; Picht et al., 2013; Sollmann et al., 2014). The task was repeated twice without stimulation to establish a baseline, from which all objects that did not elicit clear or correct responses were excluded. The baseline objects were then utilized under rTMS for the mapping phase (Krieg et al., 2016; Picht et al., 2013; Sollmann et al., 2014).

108

L. Tussis et al. / Brain & Language 168 (2017) 106–116

During stimulation, the coil was moved over 46 standardized stimulation targets that had been manually located on the individual MRI data prior to the mappings. The Cortical Parcellation System (CPS) was used for better localization and standardization of the stimulation targets (Corina et al., 2005, 2010; Krieg et al., 2016; Sollmann et al., 2014) (Fig. 1, supplementary table 1). The cortical areas that were not targeted during rTMS were the aITG, aMFG, aMTG, aSFG, dLOG, mITG, orIFG, pITG, polIFG, polITG, polLOG, polMFG, polMTG, polSFG, polSTG, and vLOG. The anatomical abbreviations of the CPS and the corresponding stimulation spots are depicted in the supplementary material. Overall, the 46 predefined cortical spots were stimulated twice, and each spot was stimulated 3 times in a row, leading to a total amount of 6 stimulations per target with a total number of (2  3  46) = 276 stimulations applied to the LH of each subject. After the first round of mapping, which included stimulation of all of the 46 points (3 stimulations per target), a short break of 5–10 min was established for recovery, and the second round of stimulation was performed subsequently, again stimulating all of the 46 points (3 stimulations per target). During the IPI and after 3 stimulations over the same spot, the magnetic coil was moved to one of the next, yet unstimulated targets. In this context, rTMS routinely started within Broca’s area, subsequently followed by stimulation of adjacent cortical regions in an undulated order. No specific order or systematic randomization of the stimulation course was established. The same MRI datasets and stimulation points were used for the L1 and L2 mapping sessions. Furthermore, at the end of both mappings, the subjects were asked to rate how much pain they felt according to the Visual Analogue Scale (VAS) (Langley & Sheppeard, 1985; Price, McGrath, Rafii, & Buckingham, 1983), overall and solely at the temporal stimulation spots. 2.6. Video and post hoc analysis After language mapping, the recordings by the camera were analyzed, in line with previous reports (Hernandez-Pavon et al., 2014; Krieg et al., 2016; Lioumis et al., 2012; Picht et al., 2013; Sollmann et al., 2014). Both video and audio data were available, but no voice-latency measurements were assessed. For each participant, three different datasets were recorded, which included task performance during the first and second baseline testing as well as task performance during rTMS mapping. When the video was played, the software automatically depicted the object belonging to the respective naming response at the bottom of the screen. In addition, baseline performance regarding the respective object was available simultaneously. Accordingly, the baseline and mapping phase were compared to identify changes in task performance due to stimulation. In the case that a naming error was identified, recording was stopped, and a tag, including definition of the type of error, was placed in the dataset. Based on these tags, the software automatically created a list of errors, which also included the corresponding stimulation burst details for post hoc assessment of the regions that were prone to errors. All of the video material was carefully analyzed at least twice in a row to reduce intraobserver variability, and a trained linguist was available to evaluate unclear cases. The errors were classified as no response, performance error, hesitation, semantic paraphasia, phonological paraphasia, or neologism (Corina et al., 2010; Krieg et al., 2016; Lioumis et al., 2012; Sollmann et al., 2014). For the non-German language trial, the subjects helped in the video evaluation to ensure that a native speaker analyzed the data. With respect to the error types and analysis procedure, each volunteer was precisely instructed before the data evaluation. Additionally, an exemplary video dataset, which was derived from rTMS-based mapping in one member of our research group, was provided as

training material. This video was approximately 15 min long, and it contained naming errors of the different categories as well as stimulations that did not elicit errors. Importantly, the stimulation protocol and task were the same during the mapping performed to generate training material when compared to the settings of the present study. First, the analysis of the training dataset was performed together with the investigators, and a second round of video evaluation was carried out by the volunteers without the investigators’ presence. This second evaluation was then checked and discussed, and analysis of video data of the present study was started afterwards. In this context, video material of the present study was always analyzed by both the respective volunteer and the investigator together. Based on the naming errors observed, we calculated the error rates (ERs) by dividing the error number by the stimulation trials for each category. Moreover, we divided the number of subjects who had committed an error by the number of subjects who underwent stimulation for each naming error category. We projected the results onto the CPS in order to visualize and compare the cortical distributions of the ERs (Corina et al., 2005, 2010) (Fig. 1, supplementary table 1). 2.7. Statistics Data analyses were conducted using GraphPad Prism (version 6.0, GraphPad Software Inc., La Jolla, CA, USA). For all statistical testing, the statistical significances were set at p < 0.05. A paired t-test was used to investigate the difference between the L1 and L2 baseline values, and a Wilcoxon matched-pair test was used to compare mean ERs for L1 and L2 for each stimulation site. As a measure for effect sizes, r was additionally calculated and reported. 3. Results 3.1. Subject and mapping characteristics Table 1 provides the subject-related characteristics, including their L1, L2, age, discomfort perceived during rTMS (according to the VAS), and age of L2 acquisition. The RMT was 36.0 ± 6.2% of the system’s maximum output for the L1 mapping and 37.1 ± 6.6% of the system’s maximum output for the L2 mapping (p = 0.54). No adverse events were observed. 3.2. Baseline All of the subjects underwent baseline testing twice. During the assessment for the L1, the subjects were able to correctly name 93.8 ± 8.0 out of 131 objects (range: 80–108 objects), whereas they correctly named 97.1 ± 10.3 objects during the L2 baseline testing (range: 81–112 objects). In this context, none of the volunteers showed a difference of more than 13 correctly named baseline objects (10% of the overall amount of presented objects) between L1 and L2 testing, leading to no subject exclusion due to differences in language proficiency in relation to baseline testing. Furthermore the difference between L1 and L2 baselines was not statistically significant (p = 0.27). 3.3. Error rates Overall, rTMS led to a mean ER of 20.8 ± 12.8% in the L1 and 13.6 ± 6.4% in the L2 for all naming errors together (Fig. 2). Furthermore, the average ER of no-response errors was 11.9 ± 9.0% in the L1 and 6.5 ± 5.2% in the L2, whereas stimulation resulted in a mean ER of 14.0 ± 10.1% for the L1 and 8.5 ± 5.5% for the L2 regarding all

109

L. Tussis et al. / Brain & Language 168 (2017) 106–116

Fig. 1. Cortical Parcellation System and stimulation targets. Template showing the locations of the 46 predefined cortical spots within the left hemisphere (LH). Each cortical target was stimulated 6 times throughout the experiment.

Table 1 Subject details and stimulation parameters. The table shows the subjects’ details, including their first language (L1) and second language (L2), age, pain or discomfort during stimulation according to the Visual Analogue Scale (VAS), and age of L2 acquisition. SD = standard deviation, Min = minimum, Max = maximum. Volunteer

1 2 3 4 5 6 7 8 9 10 Mean SD Median Min Max

Languages (L1 & L2)

Age (years)

L1

L2

Italian German Slovakian Chinese Slovakian Chinese English French Luxemburgish Luxemburgish – – – – –

German Italian German German German German German Luxemburgish Cantonese German – – – – –

23 27 19 25 25 23 24 22 23 23 23.4 2.1 23 19 27

errors without hesitation (Fig. 2). Accordingly, there was a statistically significant difference in ERs between the L1 and L2 for noresponse errors (p = 0.03), but not for all errors (p = 0.06) or all errors without hesitation (p = 0.08), respectively (Fig. 2). In addition, a comparatively large effect size was observed in terms of no-response errors (r = 0.61) and all errors together (r = 0.60), whereas a medium effect size was registered for all errors without hesitation (r = 0.45). 3.4. Error distribution per stimulation point Overall, for the L1, more areas were prone to any type of error when compared to the L2 (Figs. 3 and 4). Furthermore, the L1 showed a greater overall error distribution, with very high ERs within the mPoG (32%), mMFG (32%), pMTG (28%), pSMG (28%), and mSFG (28%; Fig. 3a). The L2, on the other hand, had its peaks at stimulation points within the anG (23% and 22%), trIFG (22%), and mPrG (22%; Fig. 3b). When not considering hesitation errors, the pMTG (25%) and mMTG (23%) showed the highest ERs for the L1 (Fig. 3c), whereas no particular pattern was observed for the L2, which had its ER peaks within the trIFG (15%), mPrG (15%), and anG (15%; Fig. 3d). No-response errors were elicited most fre-

Pain during stimulation Temporal

Convexity

5 3 7 4 3 8 3 0 2 3 3.8 2.3 3 0 8

2 1 6 3 1 6 2 0 4 2 2.7 2.1 2 0 6

Age of L2 acquisition (years)

0 3 5 5 10 6 2 3 0 0 3.4 3.2 3 0 10

quently when the pSMG was stimulated (22%) for the L1 (Fig. 3e), whereas the ER peak was reached by the trIFG for the L2 (13%; Fig. 3f). For all errors together, there was a statistically significant difference between the ERs of the L1 and L2 for stimulation spot 31 (p < 0.05, Figs. 1 and 4), but considerable differences were also detected for stimulation spots 7, 32, and 40 (p < 0.1, Figs. 1 and 4). Regarding all errors without hesitation, a statistically significant difference was observed for stimulation spots 5 and 40 (p < 0.05, Figs. 1 and 4), but targets 31 and 42 were also characterized by an above-average difference in ERs between languages (p < 0.1, Figs. 1 and 4). For no-response errors, there was a statistically significant difference for stimulation spots 17 and 42 (p < 0.05, Figs. 1 and 4), but targets 24 and 44 also showed a considerable difference in ERs between the L1 and L2 (p < 0.1, Figs. 1 and 4). 3.5. Error distribution per subject Overall, the templates visualizing total error distribution per volunteers show that more subjects committed errors in their L1 than in their L2 (Fig. 5). For all errors in the L1, ERs of at least

110

L. Tussis et al. / Brain & Language 168 (2017) 106–116

4.1. Cortical distribution of language

Fig. 2. Error rates. The figure shows the error rates (ERs) in percentages for all errors (L1 mean 20.8 ± 12.8%, L2 mean 13.6 ± 6.4%; p = 0.06), no response (L1 mean 11.9 ± 9.0%, L2 mean 6.5 ± 5.2%; p = 0.03), and all errors without hesitation (L1 mean 14.0 ± 10.1%, L2 mean 8.5 ± 5.5%; p = 0.08) for the first (L1) and second language (L2) in bilinguals. The ERs were obtained by dividing the number of errors by the number of stimulations.

80% occurred after rTMS to various regions, including the pMFG (100% and 80%), mMFG (80%), pSFG (80%), trIFG (80%), vPrG (80%), aSTG (80%), pSMG (80%), pMTG (80%), and anG (80%; Fig. 5a). In the L2, the highest ERs were detected within the pMFG (80%), mMFG (80%), and anG (80%; Fig. 5b). When looking at all errors without hesitation, rTMS to a stimulation spot within the pMFG elicited an error in 90% of the subjects for the L1 (Fig. 5c). For the L2, on the other hand, no specific area induced a large percentage of volunteers to commit a naming error; the maximum observed was 50% of the subjects at the pMFG, trIFG, aSTG, mSFG, and vPrG (Fig. 5d). Concerning no-response errors committed when investigating the L1, this error category was elicited in 80% of subjects when the pMFG was stimulated, and in 70% of volunteers with regard to the mPrG, pMTG, and mMTG (Fig. 5e). For the L2, fewer subjects committed no-response errors when stimulated, with a peak at 50% of volunteers for the mPrG, aSTG, and anG (Fig. 5f).

4. Discussion It is not fully understood whether the L1 and L2 are processed by the same cortical areas or by different ones. Currently, research suggests that the L1 and L2 are partially coded by shared cortical areas and partially by areas that are specific to either the L1 or L2. Because the majority of past studies used neuroimaging techniques, which primarily investigate correlational relationships, the data do not primarily show causal relationships. The present pilot study used rTMS—a non-invasive technique that investigates causal links between variables—to examine which areas are functional to language processing in the LH for the L1 and L2 during an object-naming task. We hypothesized that the L1 and L2 do not entirely present the same cortical distribution within the LH in the context of the assessment of an object-naming task, and that the L1 presents a more consistent pattern across the participants, whereas the L2 is distributed more heterogeneously across the subjects.

The results showed that the baseline values did not significantly differ across the L1 and L2. Since the baseline was the number of objects that the respective subject was able to name correctly and without delay, we expected that a higher baseline would be correlated to a greater knowledge of the specific language. Since the numbers of objects that constituted the baselines for the L1 and L2 did not differ significantly, proficiency in the L1 and L2 could be regarded as comparable; at least with respect to the objects shown during the naming task. Given the similarity in the L1 and L2 proficiency of our subjects, the major attribution of the results to a difference in language mastery becomes subordinate. According to the mapping results, the ERs differed clearly between the L1 and L2. Specifically, under stimulation, the LH was prone to more language errors when the object-naming task was conducted in the L1 than in the L2 (Figs. 3 and 4). Hence, the LH seems to be serving a greater role in language processing for the L1 compared to the L2 during object naming, or, likewise, the L1 can be disrupted more easily by rTMS than the L2. Against this background, the areas that gave rise to the most naming errors for the L1 and L2 only partially overlapped (Figs. 3 and 4), which suggests that L1 and L2 representations partially differ in cortical distribution, thus confirming our first hypothesis. This finding is principally in line with previous literature, which reported on language-specific, but also shared cortical regions for the L1 and L2 (Calabrese et al., 2001; Chee et al., 1999, 2001; Dehaene et al., 1997; Illes et al., 1999; Lucas et al., 2004; Pouratian et al., 2000; Roux & Tremoulet, 2002). However, previous findings seem to be highly dependent on the applied task, the age of L2 acquisition, language proficiency, and other factors (Blom et al., 2014; Buckner et al., 1995), which makes it difficult to draw final conclusions about the exact localization of language-specific and shared regions in bilinguals. In the present study, language-specific areas were observed within the frontal, parietal, and temporal lobes, with a trend towards higher occurrence of language-specific stimulation spots within posterior regions (Fig. 4). More detailed, statistically significant differences between task performances in the L1 and L2 were observed for stimulation spots 5, 17, 31, 40, and 42, whereas further areas showed non-significant trends, depending on the error category (Fig. 4). In general, this distribution seems to be in comparatively good accordance with previous studies using intraoperative stimulation among bilinguals, which reported on language-specific sites within posterior inferior parietal and temporo-parietal regions (Lucas et al., 2004; Roux & Tremoulet, 2002). However, it is important to note that direct comparison of results achieved during widespread mapping by rTMS or fMRI to intraoperative stimulation results is restricted to a comparatively low amount of areas since intraoperative stimulation is clearly limited by the borders of craniotomy. When looking at all errors together, the L1 showed a very wide distribution of areas that were prone to a significant number of errors (Figs. 3 and 4). On the other hand, the areas that resulted in the most errors for the L2 appeared to be more circumscribed, with ER peaks within the trIFG, mPrG, and anG (Figs. 3 and 4). However, the all-error category also includes hesitation errors. This type of error is not as easily classifiable as the rest of the error categories without latency measurements, for instance. Therefore, mistakes in object naming could be classified wrongly due to their comparatively unclear nature. In this context, it is not completely clear how to judge hesitations elicited by rTMS: on the one hand, they could be related to a too-sensitive examiner, but, on the other hand, they could also reflect a mild form of a no-response error. Hence, hesitations were not always explicitly included in the anal-

L. Tussis et al. / Brain & Language 168 (2017) 106–116

111

Fig. 3. Error distribution per stimulation point. The figure shows the error rates (ERs) for all of the subjects when stimulated on the cortical spots for all errors L1 (a), all errors L2 (b), all errors without hesitation L1 (c), all errors without hesitation L2 (d), no response L1 (e), and no response L2 (f). L1 = first language, L2 = second language.

ysis (Corina et al., 2010; Lioumis et al., 2012; Rosler et al., 2014). However, since a considerably large number of hesitations was elicited in the present study and in previous approaches (Krieg et al., 2016; Sollmann et al., 2014), it becomes likely that this error category is actually a result of naming disruption, suggesting that it does not solely represent a mild form of a no-response error but rather constitutes a separate type (Indefrey, 2011). However, since latency measurements were not applied in the present study and the role of hesitations in rTMS language mapping is contestable, the error distribution of all errors without hesitation might potentially provide more reliable data. When excluding hesitation errors, the L1 still presents a wider cortical distribution of language areas compared to the L2 (Figs. 3 and 4). Most importantly, the error distribution per subject

reveals the pMFG to be the area that elicited language errors in the majority of participants, with a percentage of 90% (Fig. 5). In this context, the error distribution per subject gives important information on the reliability of the data from the error distribution per stimulation point. Moreover, it allows for the identification of outliers. If rTMS to a certain stimulation point elicited errors in a small percentage of participants, this finding could not be regarded as a good indicator that an area is most probably essential for language because a majority of the volunteers did not commit errors. If, on the other hand, a stimulation point is characterized by a high percentage of subjects with errors, then that area is more likely to have an important function in language processing. As such, our data suggest that, within the dominant LH, the pMFG is involved in L1 processing to an above-average extent in bilinguals. The L2,

112

L. Tussis et al. / Brain & Language 168 (2017) 106–116

Fig. 4. Comparison of error rates at stimulation targets. This figure compares the error rates (ERs) that occurred during stimulation at the different predefined stimulation targets for the first (L1) and second language (L2). Differences between ERs of the L1 and L2 were assessed by Wilcoxon matched-pair tests for each stimulation spot. (a) shows the results for all errors, whereas (b) depicts the comparison for all errors without hesitation and (c) illustrates error distributions for no responses. * = p < 0.1, ** = p < 0.05.

on the other hand, does not present areas that were prone to a high number of errors on first glance, at least when excluding hesitations. When looking at no-response errors, once more the L1 and L2 differed in the distribution of language areas, and both presented

more circumscribed error distributions (Figs. 3 and 4). In general, with respect to the potential causes of no responses in anterior brain regions, such as the trIFG, opIFG or vPrG, they are likely correlated to an interference with speech-motor commands (vPrG) or articulatory planning (in the trIFG, opIFG) (Pouratian &

L. Tussis et al. / Brain & Language 168 (2017) 106–116

113

Fig. 5. Error distribution per subject. This figure illustrates the percentage of subjects that committed a naming error under stimulation on the specific stimulation targets. (a) shows the subject percentage for all errors L1, (b) for all errors L2, (c) for all errors without hesitation L1, (d) for all errors without hesitation L2, (e) for no response L1, and (f) for no response L2. L1 = first language, L2 = second language.

Bookheimer, 2010; Price, 2000). Concerning posterior regions such as the aSMG, pSMG or anG, the no-response errors might have primarily been due to anomic aphasia. In this context, patients with lesions located within these posterior regions can suffer from anomic aphasia, and this observation might be comparable to the no responses due to the supposed disruption of the semantic system’s output (Raymer et al., 1997). Interestingly, the area that plays a major role in L2 processing, according to our pilot study, is the trIFG, among others, which

has been shown to be actively involved in language functions in previous research on Broca’s area (Broca, 1861). Moreover, the error distribution per subject is wider for the L2 than for the L1, and this pattern is observed across all errors, all errors without hesitation, and no-response errors (Figs. 3 and 5). In fact, the L1 has specific areas that were consistently prone to naming errors in a high percentage of participants, such as the pMFG for all errors without hesitation, and the pMFG, mPrG, mMTG, and pMTG for noresponse errors (Figs. 3 and 5). On the other hand, the L2 did not

114

L. Tussis et al. / Brain & Language 168 (2017) 106–116

present areas that were prone to naming errors in a sustained number of participants; however, the subjects committed errors that were more widely distributed (Figs. 3 and 5). This seems to confirm our second hypothesis. Correspondingly, the L1 presented a more homogeneous pattern across the volunteers, reflecting a more stable pattern of language distribution. This is an interesting factor that does not seem to be dependent on the language itself. In fact, the participants had different L1 but still maintained similarities in their distributions. This suggests that the acquisition of the L1 might result in the development of the same language-related areas, independently of the specific language. Furthermore, the L2 did not show areas that were reliably prone to naming errors across the participants in our pilot study. Accordingly, the distribution is wider across the LH, which suggests that the pattern of the L2 is more heterogeneous (Figs. 3 and 4). However, this is not necessarily derived from the observation that the L2 is coded by different regions from person to person. Hence, despite baseline values suggested comparable fluency in the two languages, another important factor that could have contributed to a heterogeneous pattern was the age of language acquisition. In this paper, we adopted Kohnert’s definition of bilingualism, as no official one has been agreed upon to our knowledge (Kohnert, 2010). However, the stages of L2 acquisition could have influenced its distribution at the cortical level differently (Bloch et al., 2009; Klein, Mok, Chen, & Watkins, 2014). Specifically, Bloch et al. (2009) tested the hypothesis that the age of L2 acquisition influenced cortical language representation in a group of multilinguals with different ages of exposure to L2 who were all fluent in a latelearned other language (Bloch et al., 2009). Interestingly, the authors were able to demonstrate that regional activation during a language-production task was subject to high interindividual variability, which was higher than the observable intraindividual variability regarding the different languages (Bloch et al., 2009). Furthermore, they showed that later stages of L2 acquisition influenced the distribution across the cortical areas, favoring a higher variability across the subjects (Bloch et al., 2009). On the contrary, if the L2 is acquired at early stages, its distribution has a more stable configuration across bilingual subjects (Bloch et al., 2009), suggesting that the age of language learning has a core impact on brain plasticity and organization. This might also be reflected by our results showing variability in the mapping results between volunteers. In addition to these findings, Klein et al. measured the cortical thickness of 22 monolinguals and 66 bilinguals through MRI analysis; the enrolled bilinguals had learned both languages simultaneously or during either early (4–7 years) or old (8–13 years) childhood, respectively (Klein et al., 2014). In short, the authors were able to demonstrate that the cortical distribution of the L2 is dependent on the age at which the language is acquired, because later acquisition of the L2 was correlated to a significantly thicker cortex in the left inferior frontal gyrus and a thinner cortex in the right inferior frontal gyrus (Klein et al., 2014). Moreover, they came to the conclusion that learning the L2 after L1 modifies the cortical structure in an age-dependent way, whereas simultaneous acquisition of the L1 and L2 does not seem to lead to significant structural modification of the cortex (Klein et al., 2014). Both the functional (Bloch et al., 2009) and structural (Klein et al., 2014) findings show that the acquisition of more than one language clearly influences the cortical state, and rTMS language mapping could become a promising tool for further investigating this topic in larger and more homogeneous cohorts in the future. Furthermore, the combination of rTMS language mapping with other techniques like diffusion tensor imaging fiber tracking (DTI FT), for instance, might provide additional insights regarding the subcortical architecture of bilinguals (Sollmann, Giglhuber, Tussis, et al., 2015; Sollmann et al., 2016).

4.2. Limitations Although our pilot study provides data that show a different cortical distribution for the L1 and L2, some considerations have to be made on the study’s limitations. First, the sample size of the study was comparatively small. Consequently, a larger sample size could strengthen and refine our results. Moreover, as mentioned before, the subjects varied in the age of L2 acquisition. Correspondingly, some individuals acquired their L2 in very early childhood, whereas others did at a later stage. This variation in L2 acquisition may have influenced the results, as different ages of language acquisition could potentially have resulted in the involvement of different cortical areas (Klein et al., 2014). In this context, proficiency in the L1 compared to L2 was primarily assessed by self-report or differences in baseline testing, which could be improved by adding standardized tests for more elaborate evaluations of language proficiency in upcoming studies. In addition, considering recent models of language processing as a network function, our cortical mapping approach does not explicitly account for this modern hodotopical view (De Benedictis & Duffau, 2011; Duffau, Moritz-Gasser, & Mandonnet, 2014). In this context, the classical view of a rigid and strictly predetermined organization pattern in the brain regarding the representation of specific cognitive functions becomes more and more obsolete, and a more dynamic organization of anatomo-functional regions becomes relevant in the scientific and clinical setting. Correspondingly, in the hodotopical model, brain functions are the result of parallel streams of information that are dynamically modulated within an interactive circuit (De Benedictis & Duffau, 2011). This framework can principally be studied through a combination of different modalities capturing different anatomo-functional aspects of a respective function, such as fMRI or DTI FT, for instance. Although cortical mapping of language distribution was in the focus of the present approach, rTMS mapping data can be used for further subcortical exploration within the context of rTMS-based DTI FT (Sollmann, Giglhuber, Tussis, et al., 2015; Sollmann et al., 2016). Concerning the stimulation protocol, we used 100% of the individual RMT in combination with a stimulation frequency of 5 Hz/5 pulses for rTMS-based mapping. However, it has been demonstrated repeatedly that alterations in coil orientation, stimulation intensity, or stimulation frequency might lead to different results regarding the distribution of naming errors during task performance (Epstein et al., 1996; Hauck et al., 2015; Pascual-Leone, Gates, & Dhuna, 1991; Sollmann, Ille, Obermueller, et al., 2015). Hence, it is important to state that a considerably large number of mapping protocols can be applied in principle, which might result in different error maps. Against this background, we decided to standardly stimulate with 100% RMT and 5 Hz/5 pulses since this protocol is among the most common ones and has repeatedly shown to be easily tolerable and efficient in eliciting naming errors during object-naming tasks (Krieg et al., 2016; Rosler et al., 2014; Sollmann et al., 2014; Tarapore et al., 2013). However, we do not claim that this protocol is the most optimal one, and further studies on protocol comparisons within the scope of rTMS mappings are mandatory. Moreover, rTMS language mapping was shown to have a high sensitivity in combination with a low specificity, meaning that with rTMS, we are rather able to map languageinvolved but not necessarily language-eloquent brain regions (Picht et al., 2013). 4.3. Significance of the study To our knowledge, this is the first study to systematically investigate cortical language distribution in bilinguals by a rigorous rTMS protocol. Thus, our results should not be regarded as a final

L. Tussis et al. / Brain & Language 168 (2017) 106–116

investigation revealing all aspects of bilingual cortical language function and distribution by the use of rTMS, but rather as a pilot study showing significant differences as a basis for further and more specific questions. In this context, whether the L1 and L2 are coded by the same areas or by different ones has been a topic of great interest. As mentioned before, a majority of studies have suggested that the L1 and L2 are partially coded by the same areas but that they also present language-specific areas (Calabrese et al., 2001; Chee et al., 1999, 2001; Dehaene et al., 1997; Illes et al., 1999; Lucas et al., 2004; Pouratian et al., 2000; Roux & Tremoulet, 2002). Most of these and other studies used fMRI to explore language function, whereas our study applied rTMS to investigate the cortical distribution of the L1 and L2. Regarding the results, our study supports the theory that the L1 and L2 are partially coded by different brain areas, confirming our first hypothesis. In this context, this is the first study to highlight which cerebral structures might be functional for the L1 and L2 by the use of rTMS. Specifically, the L1 seems to be predominantly coded by the pMFG, vPrG, and the pMTG, compared to the trIFG for the L2. Moreover, the L1 and L2 were also shown to differ in their patterns of language distribution across the subjects. The L1 areas were more similar across the volunteers, whereas the patterns of language areas varied greatly for the L2. To further explore the subcortical structure of language pathways in bilinguals, the presented rTMS mapping data could be used for multimodal approaches such as rTMS-based DTI FT, for example (Sollmann, Giglhuber, Tussis, et al., 2015; Sollmann et al., 2016). In addition, task-fMRI could be conducted to distinctly explore both local activity and dynamic functional connectivity during task performance in the context of regional variability regarding L1 and L2 distributions. If carried out in the same cohort of volunteers, areas characterized by high ERs could be systematically compared to fMRI activation patterns or used as seed regions for seed-based connectivity analyses. Nevertheless, future studies that include more subjects and a more homogeneous cohort regarding the age of L2 acquisition are crucial to draw final conclusions about language distribution in bilinguals derived from rTMS mapping.

5. Conclusions This pilot study is one of the first rTMS-based studies to investigate the cortical distribution of the L1 and L2 in the LHs of healthy bilinguals during performance of an object-naming task. In line with our hypotheses described initially, our study provides evidence that the cortical language distribution of bilinguals partially differs between their L1 and L2 with language-specific areas being predominantly located in posterior brain regions. Moreover, it also provides evidence for a more stable distribution of language areas for the L1 and a more variant spatial distribution for the L2. However, since the size of the cohort was comparatively small, our pilot study intends to share the method and first data with the neuroscientific community, and it should be followed by further studies to refine the presented results.

Disclosure SK is a consultant for BrainLab AG (Munich, Germany) and Nexstim Plc. (Helsinki, Finland). The study was completely financed by institutional grants from the Department of Neurosurgery and the Section of Neuroradiology. All of the authors report that they have no conflicts of interest regarding the present study.

115

Acknowledgments NS gratefully acknowledges the support of our Graduate School’s Faculty of Medicine Graduate Center. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.bandl.2017.01. 011. References Bloch, C., Kaiser, A., Kuenzli, E., Zappatore, D., Haller, S., Franceschini, R., et al. (2009). The age of second language acquisition determines the variability in activation elicited by narration in three languages in Broca’s and Wernicke’s area. Neuropsychologia, 47(3), 625–633. Blom, E., Kuntay, A. C., Messer, M., Verhagen, J., & Leseman, P. (2014). The benefits of being bilingual: Working memory in bilingual Turkish-Dutch children. Journal of Experimental Child Psychology, 128, 105–119. Broca, P. (1861). Remarks on the seat of the faculty of articulated language, following an observation of aphemia (loss of speech). Bulletin de la Société Anatomique, 6, 330–357. Buckner, R. L., Raichle, M. E., & Petersen, S. E. (1995). Dissociation of human prefrontal cortical areas across different speech production tasks and gender groups. Journal of Neurophysiology, 74(5), 2163–2173. Calabrese, P., Neufeld, H., Falk, A., Markowitsch, H. J., Muller, C., Heuser, L., et al. (2001). Word generation in bilinguals–fMRI study with implications for language and memory processes. Fortschritte der Neurologie-Psychiatrie, 69(1), 42–49. Chee, M. W., Hon, N., Lee, H. L., & Soon, C. S. (2001). Relative language proficiency modulates BOLD signal change when bilinguals perform semantic judgments. Blood oxygen level dependent. Neuroimage, 13(6 Pt 1), 1155–1163. Chee, M. W., Tan, E. W., & Thiel, T. (1999). Mandarin and English single word processing studied with functional magnetic resonance imaging. Journal of Neuroscience, 19(8), 3050–3056. Corina, D. P., Gibson, E. K., Martin, R., Poliakov, A., Brinkley, J., & Ojemann, G. A. (2005). Dissociation of action and object naming: Evidence from cortical stimulation mapping. Human Brain Mapping, 24(1), 1–10. Corina, D. P., Loudermilk, B. C., Detwiler, L., Martin, R. F., Brinkley, J. F., & Ojemann, G. (2010). Analysis of naming errors during cortical stimulation mapping: Implications for models of language representation. Brain and Language, 115(2), 101–112. De Benedictis, A., & Duffau, H. (2011). Brain hodotopy: From esoteric concept to practical surgical applications. Neurosurgery, 68(6), 1709–1723. discussion 1723. Dehaene, S., Dupoux, E., Mehler, J., Cohen, L., Paulesu, E., Perani, D., et al. (1997). Anatomical variability in the cortical representation of first and second language. NeuroReport, 8(17), 3809–3815. Duffau, H., Moritz-Gasser, S., & Mandonnet, E. (2014). A re-examination of neural basis of language processing: Proposal of a dynamic hodotopical model from data provided by brain stimulation mapping during picture naming. Brain and Language, 131, 1–10. Epstein, C. M., Lah, J. J., Meador, K., Weissman, J. D., Gaitan, L. E., & Dihenia, B. (1996). Optimum stimulus parameters for lateralized suppression of speech with magnetic brain stimulation. Neurology, 47(6), 1590–1593. Hauck, T., Tanigawa, N., Probst, M., Wohlschlaeger, A., Ille, S., Sollmann, N., et al. (2015). Stimulation frequency determines the distribution of language positive cortical regions during navigated transcranial magnetic brain stimulation. BMC Neuroscience, 16(1), 5. Hernandez-Pavon, J. C., Makela, N., Lehtinen, H., Lioumis, P., & Makela, J. P. (2014). Effects of navigated TMS on object and action naming. Frontiers in Human Neuroscience, 8, 660. Illes, J., Francis, W. S., Desmond, J. E., Gabrieli, J. D., Glover, G. H., Poldrack, R., et al. (1999). Convergent cortical representation of semantic processing in bilinguals. Brain and Language, 70(3), 347–363. Ilmoniemi, R. J., Ruohonen, J., & Karhu, J. (1999). Transcranial magnetic stimulation– A new tool for functional imaging of the brain. Critical Reviews in Biomedical Engineering, 27(3–5), 241–284. Indefrey, P. (2011). The spatial and temporal signatures of word production components: A critical update. Frontiers in Psychology, 2, 255. Klein, D., Mok, K., Chen, J. K., & Watkins, K. E. (2014). Age of language learning shapes brain structure: A cortical thickness study of bilingual and monolingual individuals. Brain and Language, 131, 20–24. Kohnert, K. (2010). Bilingual children with primary language impairment: Issues, evidence and implications for clinical actions. Journal of Communication Disorders, 43(6), 456–473. Krieg, S. M., Shiban, E., Buchmann, N., Gempt, J., Foerschler, A., Meyer, B., et al. (2012). Utility of presurgical navigated transcranial magnetic brain stimulation for the resection of tumors in eloquent motor areas. Journal of Neurosurgery, 116 (5), 994–1001.

116

L. Tussis et al. / Brain & Language 168 (2017) 106–116

Krieg, S. M., Sollmann, N., Tanigawa, N., Foerschler, A., Meyer, B., & Ringel, F. (2016). Cortical distribution of speech and language errors investigated by visual object naming and navigated transcranial magnetic stimulation. Brain Structure & Function, 221(4), 2259–2286. Krieg, S. M., Tarapore, P. E., Picht, T., Tanigawa, N., Houde, J., Sollmann, N., et al. (2014). Optimal timing of pulse onset for language mapping with navigated repetitive transcranial magnetic stimulation. Neuroimage, 100, 219–236. Langley, G. B., & Sheppeard, H. (1985). The visual analogue scale: Its use in pain measurement. Rheumatology International, 5(4), 145–148. Lioumis, P., Zhdanov, A., Makela, N., Lehtinen, H., Wilenius, J., Neuvonen, T., et al. (2012). A novel approach for documenting naming errors induced by navigated transcranial magnetic stimulation. Journal of Neuroscience Methods, 204(2), 349–354. Lucas, T. H., 2nd, McKhann, G. M., 2nd, & Ojemann, G. A. (2004). Functional separation of languages in the bilingual brain: A comparison of electrical stimulation language mapping in 25 bilingual patients and 117 monolingual control patients. Journal of Neurosurgery, 101(3), 449–457. Pascual-Leone, A., Gates, J. R., & Dhuna, A. (1991). Induction of speech arrest and counting errors with rapid-rate transcranial magnetic stimulation. Neurology, 41(5), 697–702. Picht, T., Krieg, S. M., Sollmann, N., Rosler, J., Niraula, B., Neuvonen, T., et al. (2013). A comparison of language mapping by preoperative navigated transcranial magnetic stimulation and direct cortical stimulation during awake surgery. Neurosurgery, 72(5), 808–819. Picht, T., Mularski, S., Kuehn, B., Vajkoczy, P., Kombos, T., & Suess, O. (2009). Navigated transcranial magnetic stimulation for preoperative functional diagnostics in brain tumor surgery. Neurosurgery, 65(Suppl. 6), 93–98. discussion 98-99. Picht, T., Strack, V., Schulz, J., Zdunczyk, A., Frey, D., Schmidt, S., et al. (2012). Assessing the functional status of the motor system in brain tumor patients using transcranial magnetic stimulation. Acta Neurochirurgica (Wien), 154(11), 2075–2081. Pouratian, N., & Bookheimer, S. Y. (2010). The reliability of neuroanatomy as a predictor of eloquence: A review. Neurosurgical Focus, 28(2), E3. Pouratian, N., Bookheimer, S. Y., O’Farrell, A. M., Sicotte, N. L., Cannestra, A. F., Becker, D., et al. (2000). Optical imaging of bilingual cortical representations. Case report. Journal of Neurosurgery, 93(4), 676–681.

Price, C. J. (2000). The anatomy of language: Contributions from functional neuroimaging. Journal of Anatomy, 197(Pt 3), 335–359. Price, D. D., McGrath, P. A., Rafii, A., & Buckingham, B. (1983). The validation of visual analogue scales as ratio scale measures for chronic and experimental pain. Pain, 17(1), 45–56. Raymer, A. M., Foundas, A. L., Maher, L. M., Greenwald, M. L., Morris, M., Rothi, L. J., et al. (1997). Cognitive neuropsychological analysis and neuroanatomic correlates in a case of acute anomia. Brain and Language, 58(1), 137–156. Rosler, J., Niraula, B., Strack, V., Zdunczyk, A., Schilt, S., Savolainen, P., et al. (2014). Language mapping in healthy volunteers and brain tumor patients with a novel navigated TMS system: Evidence of tumor-induced plasticity. Clinical Neurophysiology, 125(3), 526–536. Roux, F. E., & Tremoulet, M. (2002). Organization of language areas in bilingual patients: A cortical stimulation study. Journal of Neurosurgery, 97(4), 857–864. Ruohonen, J., & Karhu, J. (2010). Navigated transcranial magnetic stimulation. Neurophysiologie Clinique, 40(1), 7–17. Sollmann, N., Giglhuber, K., Tussis, L., Meyer, B., Ringel, F., & Krieg, S. M. (2015). NTMS-based DTI fiber tracking for language pathways correlates with language function and aphasia - A case report. Clinical Neurology and Neurosurgery, 136, 25–28. Sollmann, N., Ille, S., Obermueller, T., Negwer, C., Ringel, F., Meyer, B., et al. (2015). The impact of repetitive navigated transcranial magnetic stimulation coil positioning and stimulation parameters on human language function. European Journal of Medical Research, 20(1), 47. Sollmann, N., Negwer, C., Ille, S., Maurer, S., Hauck, T., Kirschke, J. S., et al. (2016). Feasibility of nTMS-based DTI fiber tracking of language pathways in neurosurgical patients using a fractional anisotropy threshold. Journal of Neuroscience Methods, 267, 45–54. Sollmann, N., Tanigawa, N., Ringel, F., Zimmer, C., Meyer, B., & Krieg, S. M. (2014). Language and its right-hemispheric distribution in healthy brains: An investigation by repetitive transcranial magnetic stimulation. Neuroimage, 102 (Pt. 2), 776–788. Tarapore, P. E., Findlay, A. M., Honma, S. M., Mizuiri, D., Houde, J. F., Berger, M. S., et al. (2013). Language mapping with navigated repetitive TMS: Proof of technique and validation. Neuroimage, 82, 260–272.