2.7.1. Gray Matter Intensity

Using the T1-weighted sequences co-registered to language-positive or language-negative nTMS points, the GMI was measured per stimulation point (iPlan Net server, version 3.0.1; Brainlab AG, Munich, Germany). Three pixels were randomly selected within a stimulation point and used for GMI measurements to represent the GMI per stimulation point by averaging the three values obtained (Figure 3). The mean GMI of language-positive nTMS points was recorded as mGMIp, and the mean GMI of language-negative nTMS points as mGMI<sup>n</sup> in each patient (Figure 3). Then, intensity extraction was also conducted for the cerebrospinal fluid (CSF) using three randomly selected pixels in the lateral ventricles, and the mean was recorded as mIcsf to obtain an internal control value (Figure 3). The signal intensity ratio (IR) was calculated as follows [49]:

$$IR\_p = \frac{m\text{GMI}\_p}{m\text{I}\_{\text{csf}}} \tag{1}$$

$$IR\_n = \frac{m\text{GMI}\_n}{m\text{I}\_{\text{csf}}} \tag{2}$$

**Figure 3.** Analysis of gray matter intensity (GMI). The GMI of language-positive or language-negative spots, determined by language mapping using navigated transcranial magnetic stimulation (nTMS), was measured in T1-weighted sequences. Three pixels were randomly selected within a stimulation spot and used to calculate a mean GMI for language-positive nTMS spots (mGMIp) and language-negative nTMS spots (mGMIn). The signal intensity ratio (IR) was then calculated by dividing the mGMIp or mGMI<sup>n</sup> by the mean intensity of cerebrospinal fluid (mIcsf).

### 2.7.2. U-Fibers

Using nTMS-based DTI FT with the different FAT levels and a minimum FL of 30 and 3 mm, volumes of U-fibers were measured for fiber tractography considering language-positive or language-negative nTMS points as ROIs, respectively (Brainlab Elements, version 3.1.0; Brainlab AG, Munich, Germany; Figure 4). From assessing the difference in volumes between fibers with a minimum FL of 30 mm and fibers with a minimum FL of 3 mm, the volume of U-fibers was obtained (Figure 4). In this context, U-fibers are considered short association fibers, which connect cortical regions between adjacent gyri and typically have a length in the range of 3 to 30 mm [48].

A higher number of nTMS points considered during tractography should lead to more fibers, which was corrected for by dividing the fiber volume by N<sup>p</sup> or Nn, resulting in the ratios RUfibers\_*<sup>p</sup>* and

RUfibers\_n for fibers with a length of 3 to 30 mm. The ratios in RUfibers\_*<sup>p</sup>* and RUfibers\_n were calculated as follows:

$$R\_{LIfibers\_{\!\!p\!-}} = \frac{V\_{p(3)} - V\_{p(30)}}{N\_p} \tag{3}$$

$$R\_{LIfibers\\_n} = \frac{V\_{n(3)} - V\_{n(30)}}{N\_n} \tag{4}$$

Vp(30) indicates the volume of fibers with a minimum FL of 30 mm, as derived from nTMS-based DTI FT using the ROI of language-positive nTMS points, whereas Vp(3) represents the respective volume for a minimum FL of 3 mm. Analogously, Vn(30) indicates the volume of fibers with a minimum FL of 30 mm, as derived from nTMS-based DTI FT that considers the ROI of language-negative nTMS points, whereas Vn(3) represents the respective volume for a minimum FL of 3 mm.

**Figure 4.** Analysis of subcortical fiber tracts. Tractography maps were generated based on language mapping data derived from navigated transcranialmagnetic stimulation (nTMS), using the language-positive nTMS spots and language-negative nTMS spots as separate regions of interest (ROIs). Row (**A**) visualizes the language-positive nTMS points (blue) and language-negative nTMS points (purple) in an exemplary patient case with a left-hemispheric contrast-enhancing tumor with temporo-parietal location. Row (**B**) illustrates the complete picture of tractography (green fibers) with special emphasis on U-fibers (purple fibers). Row (**C**) depicts interhemispheric fiber courses (orange fibers) crossing the midline via the corpus callosum. Row (**D**) visualizes fibers projecting to the cerebellum (orange fibers).

2.7.3. Interhemispheric Fibers and Fibers Projecting to the Cerebellum

The highest percentage of patients presenting Cross-F or Cereb-F according to visual image inspection of tractography maps was achieved for 25% FAT, which was then taken as the adjustment used for evaluation of Cross-F and Cereb-F volumes, together with a minimum FL of 30 mm derived from language-positive or language-negative nTMS points as ROIs, respectively (Brainlab Elements, version 3.1.0; Brainlab AG, Munich, Germany; Figure 4).

The volumes of the Cross-F and Cereb-F for nTMS-based DTI FT with 25% FAT were recorded as Vcross\_*<sup>p</sup>* and Vcereb\_*<sup>p</sup>* when derived from nTMS-based DTI FT using the ROI of language-positive nTMS points, whereas Vcross\_n and Vcereb\_n represent the Cross-F and Cereb-F volumes for nTMS-based DTI FT conducted with language-negative nTMS points as the ROI. The ratios Rcross\_*<sup>p</sup>* and Rcereb\_*<sup>p</sup>* and Rcross\_n and Rcereb\_n were calculated as follows:

$$R\_{\text{cross\\_}p} = \frac{V\_{\text{cross\\_}p}}{N\_p} \text{ and } R\_{\text{cerb\\_}p} = \frac{V\_{\text{cereb\\_}p}}{N\_p} \tag{5}$$

$$R\_{\text{cross\\_}l} = \frac{V\_{\text{cross\\_}n}}{N\_{\text{ll}}} \text{ and } R\_{\text{cer}b\\_ll} = \frac{V\_{\text{cereb\\_}n}}{N\_{\text{ll}}} \tag{6}$$

### *2.8. Statistical Analyses*

GraphPad Prism (version 6.04; GraphPad Software Inc., La Jolla, CA, USA) was used for statistical data analyses and generation of graphs. Descriptive statistics using relative and absolute frequencies or mean, standard deviation (SD), and ranges were calculated for demographics and characteristics of language mapping and tractography. The Shapiro–Wilk normality test was used to assess the distribution of data, which indicated a non-Gaussian distribution for the majority of data.

To investigate differences between N<sup>p</sup> and Nn, IR<sup>p</sup> and IRn, RUfibers\_*<sup>p</sup>* and RUfibers\_n (separately for nTMS-based DTI FT with 100% FAT, 75% FAT, 50% FAT, and 25% FAT), Rcross\_*<sup>p</sup>* and Rcross\_n, and Rcereb\_*<sup>p</sup>* and Rcereb\_n, Wilcoxon matched-pairs signed rank tests were performed. Furthermore, correlation analyses computing Spearman's rho were performed between Np, IRp, RUfibers\_*p*, Rcross\_*p*, and Rcereb\_*<sup>p</sup>* as well as, analogously, between Nn, IRn, RUfibers\_n, Rcross\_n, and Rcereb\_n and the status of preoperative, postoperative, and follow-up aphasia, considering the four grades as derived from language function assessments at different time points. The correlation analyses were adjusted for multiple testing using the Benjamini–Hochberg procedure with a false discovery rate of 25%.

#### **3. Results**

#### *3.1. Cohort Characteristics*

Twenty patients (mean age: 63.2 ± 12.9 years, age range: 20.3–80.8 years, 4 females and 16 males, 16 right-handers according to EHI scores) were included, all diagnosed with a left-hemispheric GBM according to histopathological evaluation (Table 1). Ten patients (50%) showed preoperative aphasia, whereas eight patients (40%) showed aphasia during follow-up examinations three months after tumor resection. GTR according to postoperative MRI was achieved in 10 patients (50%).

#### *3.2. Comparison between Language-Positive and Language-Negative Mapping and Tractography*

Language mapping of the tumor-affected LH and nTMS-based DTI FT was possible in all enrolled patients. None of the patients showed adverse events in the course of stimulation.

There were statistically significant differences in almost all measures between mapping or tractography using language-positive or language-negative nTMS points as ROIs, respectively (Table 2). In detail, patients showed a higher mean number of language-negative nTMS points (*p* = 0.0026), whereas the IR of these points was elevated on average in comparison to language-positive nTMS points (*p* = 0.0121; Table 2). The ratios for U-fiber volumes as well as for long fibers projecting to the contralateral hemisphere and fibers projecting to the cerebellum were higher for tractography using language-positive nTMS points (*p* ≤ 0.0494; Table 2).


**Table 1.**Cohort characteristics.

This table shows cohort details, including sex distribution and age (in years), information on the number of language-positive and language-negative sites according to the cortical parcellation system (CPS), the fractional anisotropy threshold (FAT) for diffusion tensor imaging fiber tracking (DTI FT) based on navigated transcranial stimulation (nTMS), and aphasia grades (at three different time points: preop = preoperative status, postop = postoperative status, follow-up = status during follow-up examinations three months after surgery). Language mapping aimed to cover 46 target points in total that were placed in relation to the CPS on the left hemisphere (LH), but a reduced number of targets was stimulated in patients with large tumor masses that precluded placement of all 46 target points. Thus, the numbers of language-positive and language-negative CPS sites do not necessarily add up to 46 in all enrolled patients.


**Table 2.** Comparison between language-positive and language-negative mapping and tractography.

This table shows the mean and standard deviation (SD) for the number (N) of language-positive and language-negative points as mapped by navigated transcranial magnetic stimulation (nTMS), intensity ratio (IR), ratio of volumes for U-fibers (RUfibers, as derived from tractography using 100%, 75%, 50%, and 25% of the individual fractional anisotropy threshold (FAT)), and ratio of volumes for interhemispheric fibers (Rcross, using tractography with 25% FAT) as well as fibers projecting to the cerebellum (Rcereb, using tractography with 25% FAT). Wilcoxon matched-pairs signed rank tests were conducted to assess differences in these characteristics between language-positive and language-negative mappings (level of statistical significance: *p* < 0.05). Statistically significant values are displayed in bold.

#### *3.3. Associations with Aphasia Grading*

For language-positive nTMS points, statistically significant positive correlations were revealed between their absolute frequency and aphasia for the preoperative (R = 0.4919, *p* = 0.0276), postoperative (R = 0.6183, *p* = 0.0037), and follow-up status (R = 0.4854, *p* = 0.0300; Table 3). The higher the number of language-positive nTMS points of the tumor-affected LH, the higher the aphasia grade. Furthermore, statistically significant negative correlations were observed between the ratio of U-fiber volumes (considering tractography with 100% FAT) and aphasia for the postoperative (R <sup>=</sup> −0.6102, *<sup>p</sup>* <sup>=</sup> 0.0043) as well as follow-up status (R <sup>=</sup> −0.4899, *<sup>p</sup>* <sup>=</sup> 0.0283; Table 3). Thus, the lower this ratio was, the higher the aphasia grade.

Regarding language-negative nTMS points, statistically significant negative correlations were revealed between their absolute frequency and aphasia for postoperative (R = <sup>−</sup>0.6097, *p* = 0.0043) and follow-up examinations (R = <sup>−</sup>0.4741, *p* = 0.0347; Table 3). Hence, the higher the number of language-negative nTMS points of the tumor-affected LH, the lower the aphasia grade.


**Table 3.** Associations with aphasia grading.


**Table 3.** *Cont.*

This table shows the correlation results between the number of language-positive and language-negative points as mapped by navigated transcranial magnetic stimulation (nTMS), intensity ratio (IR), ratio of volumes for U-fibers (RUfibers, as derived from tractography using 100%, 75%, 50%, and 25% of the individual fractional anisotropy threshold (FAT)), ratio of volumes for interhemispheric fibers (Rcross, using tractography with 25% FAT) as well as fibers projecting to the cerebellum (Rcereb, using tractography with 25% FAT) and the aphasia grades for the preoperative, postoperative, and follow-up status. Correlation coefficients are represented by Spearman's rho, and related *p*-values are given (level of statistical significance: *p* < 0.05). Statistically significant values that survived adjustments for multiple testing (Benjamini–Hochberg procedure with a false discovery rate of 25%) are depicted in bold.

#### **4. Discussion**

This study investigated the difference between language-positive and language-negative mappings as derived from presurgical nTMS and nTMS-based DTI FT in patients harboring supratentorial GBMs. There are three main results that can be taken from our analyses. First, regarding the cortical level, a significantly lower GMI was revealed for language-positive nTMS points compared to language-negative counterparts. Second, on the subcortical level, language-positive areas were characterized by an increased connectivity profile, i.e., such areas showed a significantly higher ratio in volumes for U-fibers, interhemispheric fibers, and fibers projecting to the cerebellum. Third, the number of language-positive nTMS points as well as the ratio in volumes for U-fibers were significantly associated with aphasia grading as derived from assessments at different time points.

#### *4.1. Gray Matter Intensity*

While there is evidence for alterations in GM distribution and volume related to language function, the role of the GMI to characterize language-involved areas has not been investigated to the authors' knowledge. Specifically, previous research has detected that language lateralization is predicted by the degree of GM lateralization [30]. Further, subjects diagnosed with dyslexia showed GM deficits, but the GM volume, however, can be subject to changes following training interventions in dyslexic children [31,50]. On the contrary, region-specific increases in GM volume have also been revealed for developmental language disorders, which might be interpreted as a result of compensatory mechanisms [31,32]. In the present study, lower GMI in T1-weighted sequences was revealed for language-positive nTMS points when compared to language-negative spots. This may probably reflect a sign of higher GM density, potentially suggesting increased functional involvement. Yet, this remains speculative until further studies using a similar setup can confirm our findings. For the present study, whether potentially increased functional involvement is due to higher intrinsic contribution of such areas to language function or related to compensatory mechanisms for language function at risk in our sample remains beyond the scope of investigation. However, the findings for GMI may

provide a direct link between a functionally language-related area as mapped by nTMS and structural cortical characteristics.
