Material and methods

Ethical considerations

We conducted a retrospective, single-center study, approved by the Bernese Cantonal Ethics Commission (Project ID, 2024-02510). All patients provided general consent, and the study followed STROBE guidelines.

Study population

We screened 286 patients and included 23 patients who underwent surgery between January 2019 and December 2024 (see Supplementary Figure 1). The inclusion criteria were IDH-mutant gliomas without contrast enhancement located in the left frontal lobe affecting the FAT (Louis et al., 2021). Exclusion criteria included glioblastomas, contrast-enhancing gliomas, right hemisphere tumors, and unavailable DTI tractography. This study deliberately focused on predominantly non-contrast-enhancing, IDH-mutant gliomas, as their general longer survival enables longitudinal evaluation of postoperative language recovery. The speech-eloquent hemisphere was identified using fMRI (7/23, 30.4%) or neuropsychological testing (22/23, 95.7%).

Language assessment

Language assessments were performed preoperatively (median 7.5 days before surgery), intraoperatively, postoperatively (median 2 days after surgery), and at follow-up (median 64 days after surgery). Preoperative tests evaluated spontaneous speech, auditory comprehension, object naming, and writing. Intraoperative monitoring included spontaneous speech and a picture-naming task during cortical and subcortical mapping. Only items successfully named preoperatively were used. Postoperative assessments included informal bedside screening, and follow-up assessments repeated the preoperative test battery.

Five linguistic deficit categories were defined: "language," "speech," "speech planning," "speech initiation," and "communication," based on speech-language disorder literature (Jordan and Hillis, 2006; McSween et al., 2024; Togher et al., 2023). These categories consisted of aphasia as a “language” disorder, dysarthria as a “speech” disorder and apraxia of speech as a “speech planning” disorder. Further, “communication” was defined as a cognitive communication disorder. In addition, some patients had difficulties in producing speech on their own and, in particular, in initiating utterances. The category “speech initiation” was defined for such deficits. (Fig. 2, Fig. 3).

Fig. 2: Illustration of the evolution of the distribution of deficits across five linguistic deficit categories at different time points. Scores range from 0 (no deficit) to 3 (severe deficit), with intermediate grades indicating mild and moderate impairment. The five linguistic deficit categories are defined as follows: language = aphasia as a language disorder, speech = dysarthria as a speech disorder, speech planning = apraxia of speech as a speech planning disorder, speech initiation = difficulties in producing speech, communication = pattern of cognitive communication disorder.

Fig. 3: Heatmap illustrating disruption of the FAT and AF (red, graded 0 = intact, 1 = partially disrupted, 2 = completely disrupted), and the deficits across five linguistic categories at preoperative, immediate postoperative and 3 months follow-up assessments (green). The five linguistic deficit categories are defined as follows: language = aphasia as a language disorder, speech = dysarthria as a speech disorder, speech planning = apraxia of speech as a speech planning disorder, speech initiation = difficulties in producing speech, communication = pattern of cognitive communication disorder. The shades of the colors indicate the severity of linguistic impairment with darker colors indicating more severe deficits. The illustrated data is from 19 patients, due to 4 patients with missing values for language assessment.

All language assessments were performed by two certified speech and language therapists. The categorization of linguistic deficits was based on established speech-language pathology literature. However, due to the different severity of language deficits and socioeconomic background of the individual patients, no universally accepted defined classification system for postoperative language deficits exists in this context. Thus, the categories were defined pragmatically according to the limited available evidence and expert clinical judgment.

Intraoperative neurophysiological mapping protocol

In the patients which were operated in an awake surgery setting, Penfield stimulation for mapping of language function was performed with an ISIS IOM System (Inomed®, Emmendingen, Germany) (Seidel et al., 2024). We applied square-wave biphasic current with a pulse width (PW) of 0.6 ms and a frequency of 50 Hz, using a bipolar stimulation probe with an interpolar distance of 8 mm and tip diameters of 2 mm (Inomed® REF 522 624). The stimulation current intensity was determined over the primary motor area of the tongue while the patient was counting, to check for dysarthria. This stimulation intensity, between 3 and 6 mA, was then used for further mapping of the speech eloquent cortex. During the cortical mapping, electrocorticography was recorded to detect after-discharges or spike activity with the help of subdural spider electrodes (AdTech® REF VG04A-IS00X-0KG).

For those patients undergoing intraoperative neurophysiological monitoring and mapping (IOM) under general anaesthesia, the detailed methods for motor evoked potential monitoring and continuous subcortical dynamic mapping with the monopolar suction probe are described in our previous publication (Seidel et al., 2025). The choice for awake surgery or resection under IOM was done accordingly to tumor location and patient cooperation.

Measurement of the FAT and AF

FAT and AF were reconstructed using BrainLab software (BrainLab, Munich, Germany) on pre- or postoperative DTI datasets. For FAT, the superior frontal gyrus was selected as the region of interest, connected to Broca's area. Fiber tracking parameters included FA threshold 0.15, minimum fiber length 50 mm, and maximum angulation 45°. For AF, a predefined template was used with the same parameters. (Tagliaferri et al., 2024). To assess the robustness of tract reconstruction, additional analyses were performed using different combinations of minimum fiber length, fractional anisotropy, and angulation thresholds. The 50-mm minimum fiber length provided the most anatomically plausible and reproducible reconstructions and was therefore used for the final analyses.

Classification of speech and language deficits and tract integrity

Deficits were classified into five linguistic deficit categories (see 2.3): language, speech, speech planning, speech initiation, and communication. Each was scored for severity on a four-point ordinal scale (0–3; no to severe deficit). To assess tract integrity, preoperative DTI reconstructions of the FAT and AF were fused with postoperative T2-weighted images. As postoperative DTI datasets were unavailable in the majority of cases, preoperative DTI datasets were fused with both pre- and postoperative T2-weighted imaging to evaluate whether the tumor and surgical resection cavity overlapped with or disrupted the preoperatively reconstructed FAT and AF. Tract disruption was classified as "intact" (score 0), "partial disruption" (score 1) or "total disruption" (score 2). This allowed comparison of functional deficits across domains with the structural integrity of the FAT and AF at various time points.

Statistics

For inferential statistics, we evaluated the association between each pair of ordinal scores by computing Kendall's tau correlation coefficients. Based on prior evidence and our clinical hypothesis, greater involvement of the FAT and AF was expected to be associated with more severe postoperative language deficits rather than improved language performance. We therefore tested for positive associations between tract involvement and language deficit severity using one-sided hypothesis tests. Thus, all correlation coefficients were tested using a one-sided test against the null hypothesis of no correlation. The significance level for p-values was set at 0.05. The statistical analyses were performed using R (R 4.5.0 and 4.6.1, R Core Team (2025-2026)) and RStudio (2024.12.1.563 and 2026.5.0.218, Posit team (2025-2026)).