*2.4. Language Mappings*

### 2.4.1. Setup

Before mapping started, the individual T1-weighted sequences were uploaded to the Nexstim eXimia NBS system version 4.3, (Nexstim Plc. Helsinki, Finland). Forty-six stimulation targets (placed in reference to the cortical parcellation system (CPS) [49]) were assigned to the 3D head models of each individual. They covered all cortical areas with the exception of the occipital lobe, frontal and temporal poles, and the inferior temporal regions due to inability to reach or high discomfort under stimulation. Figure 3 and Table 1 name all targeted areas. Using these predefined targets ensured the same cortical spot to be targeted in each task.

**Figure 3.** Template of the left hemisphere with 46 stimulation targets covering 21 cortical parcellation system (CPS) regions. The regions in the right hemisphere were mirror-inverted. Occipital areas, frontal and temporal poles, and inferior temporal regions were excluded due to inability of reach or high discomfort in combination with stimulation. See Table 1 for anatomical names corresponding to the abbreviations.


**Table 1.** Anatomical names and corresponding abbreviations of the 21 cortical parcellation system (CPS) regions as adapted from [49].

For nTMS, a focal figure-of-eight coil with upward handle position and automatic overheating protection was employed. It produced biphasic pulses (length: 230 µs) with a maximal electric field strength of 172 V/m ± 2% (at 25 mm depth beneath the coil in a spherical conductor model representing the head). The setup provided visualization of cortical areas to be targeted in relation to the coil's focal point as well as the e-field's orientation to the gyrus [50]. All pulse applications were tracked, controlled, and saved. Inaccurate pulse applications not on target were not saved by the software.

As a first step, the resting motor threshold (rMT) of each individual was established: surface electrodes for electromyography (EMG) were placed over the abductor pollicis brevis and abductor digiti minimi muscles. Single-pulse stimulation was applied over the anatomical hand knob to identify the most excitable spot with the electrical field perpendicular to the central sulcus. Using the built-in threshold-hunting algorithm, the lowest possible threshold to elicit at least five out of ten positive motor responses was defined as the rMT. For a more detailed description, see [18] as the most common approach used in the field as well as in the present set-up. Moreover, a rMT of 110% was employed as intensity for nTMS language mapping [18]. The rMT was defined separately for each hemisphere (for means see Table 2).

#### 2.4.2. Baseline Naming

To determine the ideal set of picture stimuli per participant, baseline naming was performed. Participants sat about 60 cm from of a screen on which the pictures were presented with a picture presentation time (PPT) of 1000 ms and an inter-picture interval (IPI) of 3000 ms. The tasks appeared in blocks per task; the item order per task was randomized.

Two rounds of baseline naming were administered, in which participants had to name the entire set of 150 pictures without stimulation. The instruction was to name pictures using the entire sentence as quickly and precisely as possible. Those picture stimuli that were not named fluently and consistently with the same label within the given time window in two rounds of baseline testing were excluded from naming under stimulation. The number of incorrectly named stimuli was documented. The procedure was audio- and video-recorded for post-hoc analysis.

#### 2.4.3. Mapping Procedure

The individualized set of stimuli resulting from the baseline naming was used per participant. Object and action naming tasks were administered in separate blocks. The task order and stimulation of hemispheres was balanced across participants, so that each task and hemisphere was targeted first in half of the participants and the respective other task and hemisphere in the other half of the participants. This was done to exclude fatigue as an influence of error rate. The same was done for stimuli order within each task: the stimuli order within each task was randomized. The individual and randomized list was used per participant and restarted, once it had reached its end during administration of each block to cover all areas.

The IPI and PPT were the same as during baseline testing. The onset of stimulation was synchronized with the onset of picture presentation; hence, the picture-to-trigger interval (PTI) was 0 ms. Stimulation consisted of 10 rTMS trains delivered at 5 Hz/5 pulses at 110% of the intensity established during motor threshold hunting.

During language mapping, the coil was placed over each of the 46 predefined stimulation target distributed over the majority of the cortical surface, while the participant named the items. Each target point was stimulated three times per round. There were two rounds per hemisphere and task, while alternating between left hemisphere (LH) and right hemisphere (RH). This means, that the overall mapping resulted in six data points per stimulation target.

The sequence of left- and right-hemispheric stimulation was randomized. The participants were instructed to report any pain and discomfort that may occur during stimulation to stop the mapping in case the procedure was intolerable.

#### 2.4.4. Mapping Analysis

Through post-hoc analysis of the recorded and segmented videos, baseline-naming performances were compared to performance under stimulation in a side-by-side comparison. The stimulation videos were screened for any of the following speech and language errors compared to the baseline counterpart. The investigator carrying out the analysis was blinded to where stimulation had taken place. Errors due to pain, discomfort, or visible stimulation of peripheral facial nerves were excluded from the analysis.

*Categories:*

Conceptual (non-linguistic) errors:


Lexico-semantic errors:


Grammatical errors:

6. Grammatical error: for example, a missing or wrong inflection for verb or noun and article.

Phonological-articulatory errors:


#### *2.5. Statistical Analysis*

All calculations were performed using R software (R Studio version 3.5.2; The R Foundation for Statistical Computing, Vienna, Austria). A *p*-value of <0.05 was considered statistically significant for all comparisons and correlations.

Baseline error rates were calculated by dividing the number of errors of the baseline by the number of items to name. To address research question 1, 3, and 4, error rates for object and action naming were calculated for the respective areas.

The error rate was defined as the number of errors divided by the total number of stimulations in a particular area. Two different overall error rates were calculated, one including all errors (categories 1–8), one without hesitation errors (excluding category 2 and 5). This was performed for errors per task, hemisphere, and CPS region in each hemisphere. Shapiro–Wilk normality tests suggested non-normal distribution of error rates. Hence, Mann–Whitney–Wilcoxon tests were conducted to assess differences in error rates between object and action naming in each of the regions. Moreover, Mann–Whitney–Wilcoxon tests were applied to compare error rates, when excluding or including hesitations, in the overall error rates.

To examine error categories regarding research question 2, error rates were calculated per category (non-linguistic speech errors, lexico-semantic errors, grammatical errors, phonological-articulatory errors) and, following the Shapiro–Wilk test, were compared between tasks per hemisphere using Mann–Whitney–Wilcoxon tests, and reported the included effect sizes. Moreover, error ratios per category were established as the errors per category divided by the overall number of errors (see Table 3 for example calculations).

Multivariate regression models were performed to evaluate the influence of baseline errors, rMT, handedness, and age on all errors, and on errors without hesitation in both hemispheres. Spearman's correlations were employed to reveal a relation between errors in the baseline and errors under stimulation for the individual tasks. Additionally, Mann– Whitney–Wilcoxon tests were used to compare error rates in the first and second round of mapping to evaluate the effect of fatigue as a potential confounding factor.

#### **3. Results**

#### *3.1. Group Characteristics and Confounding Factors*

Characteristics of the participants are summarized in Table 2. No participant had to be excluded due to pain or intolerance to nTMS or MRI acquisition; moreover, the mapping did not have to terminate early due to any such disturbance. All participants tolerated the stimulation well and reported no interference of the stimulation with the overall execution of the tasks.

**Table 2.** Demographics and error rates per round and hemisphere.


Baseline error rates for object naming amounted to 3.65 ± 2.11 and for action naming to 11.1 ± 4.424, meaning that during object naming under stimulation, 71.35 ± 2.11 remaining stimuli were used, and 63.9 ± 4.424 during action naming.

Multivariate regression models revealed that neither baseline errors (LH: t = 0.505, *p* = 0.621; RH: t = 0.522, *p* = 0.609), rMT (LH: t = 0.422, *p* = 0.679; RH: t = −0.040, *p* = 0.968), handedness (LH: t = 0.815, *p* = 0.428; RH: t = −0.231, *p* = 0.820), nor age (LH: t = −0.601, *p* = 0.557; RH: t = −0.950, *p* = 0.357) were significant predictors of the error rates in either of the hemispheres. The Mann–Whitney–Wilcoxon tests did not reveal a significant difference between the error rates of the first and second round of mapping (LH: *p* = 0.349, RH: *p* = 0.422; Table 2), nor a difference between error rates in the left and right hemisphere (*p* = 0.227).

#### *3.2. Task Comparison of All Errors*

Action naming demonstrated a significantly higher error rate than object naming in both hemispheres (LH: action naming mean error rate = 0.078, object naming mean error rate = 0.054 (*p* = 0.015, r = −0.555; RH: action naming mean error rate = 0.088, object naming mean error rate = 0.06 (*p* = 0.040, r = −0.463))). Significantly more pictures had to be excluded in the baseline naming of action naming (mean error rate = 0.124) compared to object naming (mean error rate = 0.045) (*p* < 0.001, r = 0.877), but no correlation between these error rates and the respective error rates under stimulation was found (LH: rho = 0.185, *p* = 0.435 for object naming; rho = 0.130, *p* = 0.585 for action naming; RH: rho = 0.052, *p* = 0.830 for object naming; rho = 0.339, *p* = 0.143 for action naming) for either of the tasks.

#### *3.3. Comparison of Error Categories*
