Patients with a presumed low-grade glioma (N = 15) and patients with a presumed grade I meningioma (N = 10), who were scheduled to undergo surgery at the University Medical Centre Groningen (UMCG) or the Erasmus MC University Medical Centre Rotterdam (Erasmus MC), were invited to participate in the study.

Inclusion criteria were:

Exclusion criteria were:

Glioma patients underwent tumour resection with an awake procedure and meningioma patients underwent surgery under general anaesthesia. For both aetiologies, a control group was composed of healthy volunteers (N = 15 matched to gliomas, N = 9 matched to meningiomas), matched for age, education, and gender (often proxies of the patients). The same exclusion criteria applied for these groups. This multicentre study was approved by the medical, ethical review board of the UMCG, which was also valid to the Erasmus MC. All participants gave written informed consent.

Both patient groups underwent extensive language assessment at two time points: T1 = before surgery (glioma patients: mean 33 days, range 6–83; meningioma patients: mean 20 days, range 1–48); and T2 = 12 months after surgery (glioma patients: mean 12.61, range 11.05–15.22; meningioma patients: mean 11.97, range 11.21–12.49). EEG registration was performed at T1 and at T2 in glioma patients, at T1 in meningioma patients¹, and once in healthy participants.

A selection of standardised language tests was conducted, assessing a wide range of linguistic abilities. Tests included object naming (De Witte et al., 2015), action naming in sentence context (Olde Dubbelink et al., 2013), category fluency (Szelies et al., 2002), letter fluency (Schmand et al., 2008), reading and writing (Van Ierschot, 2018), and subtests of the Diagnostic Instrument for Mild Aphasia (DIMA) (Satoer et al., 2019): repetition, semantic odd-picture-out, and sentence completion.

Raw scores were normalised by conversion to z-scores, using pre-existing normative data from a healthy population. The tests were divided over six language domains (Table 1) to cover performance in three language modalities (speech production, reading, and writing) and at three linguistic levels (phonology, semantics, and grammar). For each patient, the domain score equalled the average z-score of the tests in that domain. Domain scores below -1.5 were considered to reflect an impairment.

Electroencephalography was registered using 44-channel Schwarzer amplifiers (Natus Europe GmbH, Munich). Twenty-one scalp electrodes were applied according to the International 10–20 System, with or without a cap according to local protocol. Additional polygraphic channels were applied for artefact detection: electrooculography (EOG; two diagonally placed electrodes for eye movements), electrocardiography (ECG; chest electrode), and respiration (abdominal movements). EEG was recorded with electrode impedances ≤ 5 kΩ, a sampling frequency of 500 Hz, and with Cz as reference electrode.

At least 5 min of eyes-closed, resting-state EEG were recorded for every participant. If there were many artefacts, the duration was prolonged to 10 min. Participants were instructed to refrain from moving and to stay alert. Alertness was continuously monitored and acoustic stimuli were given when signs of drowsiness appeared. For (glioma) patients with suspected epileptiform activity, the resting-state registration was followed by the standard clinical EEG registration of at least 30 min (these data were not taken into account for our analyses).

Evaluation of slow-wave activity was performed visually,² the gold standard in clinical evaluation, and quantitatively for investigating a relation to language performance.

Visual EEG analysis was performed in BrainRT software (v.2.0; Rumst, Belgium, 2013) by one of the authors (PJC), a board-certified clinical neurophysiologist, who was blinded to the participants’ medical situation (brain tumour or healthy). The presence and characteristics of activity in the delta (0.5–4 Hz), theta (4–8 Hz), and alpha (8–12 Hz) frequency bands were determined with resting-state EEG data reformatted to display standard antero-posterior bipolar montage and common average montages. Subsequently, slow-wave activity was classified according to the Mayo Classification System (Mayo Clinic, 1998), complemented by guidelines of Lüders and Noachtar (2000). From this, four categories were created: normal, mild, moderate, and severe slow-wave activity. See Table 2 for their definitions. As pathological slow-wave activity is generally pronounced over a region of brain injury (De Jongh et al., 2003), the categories of visual EEG analysis are presented in relation to quantitatively analysed slow-wave activity over the affected area as described below.

Quantitative EEG analysis was performed in BrainVision Analyzer 2.0 (Brain Products GmbH, 2013). Artefact-free EEG (e.g., blinks/eye movements, muscle contraction artefacts) was selected by visual inspection (IB) and was segmented into 20, 2-s epochs for each participant. Raw EEG data were re-referenced to a new averaged reference consisting of 16 scalp electrodes: F3; F4; F7; F8; T3; T4; T5; T6; C3; C4; P3; P4; O1; O2; Fz; and Pz [excluding: Cz (reference electrode); Fp1 and Fp2 (to minimise excess muscle activity and eye-movement artefacts); A1 and A2 (do not register brain activity)]. Subsequently, the data were filtered (band-pass: 0.27–30 Hz, slope 24 dB/Oct; notch: 50 Hz, slope 24 dB/Oct). This type of filtering leaves a low absolute power of higher frequency bands within the filtered signal. For every patient, spectral analysis was performed on the 20 epochs of 1,000 points by a Fast Fourier Transform (FFT) for all 16 electrodes. After averaging the 20 FFTs, absolute power in the delta (0.5–4 Hz) and theta range (4–8 Hz) was normalised by dividing both by the absolute power in the total spectrum (0.5–70 Hz). Although the absolute power in the frequency range above 30 Hz will be very low, excluding the power in the assessment of the relative power could overestimate the observed relative power in the lower frequency bands. To correct for this, it was decided to use the broad band spectrum. Hence, quantitative analysis of slow-wave activity resulted in relative power values in the delta and theta band separately.

In order to examine slow-wave activity at different spatial levels over the scalp, three measures were created. First, a whole-brain measure was computed, consisting of the mean activity of the 16 remaining scalp electrodes. Subsequently, two “tumour-specific” measures were created, taking into account the tumour localisation of individual patients: (De Robles et al., 2015) “Affected hemisphere,” consisting of the mean activity of seven electrodes on the side of the tumour (F3, F7, T3, T5, C3, P3, and O1 for left-sided tumours and F4, F8, T4, T6, C4, P4, and O2 for right-sided tumours); and (Davie et al., 2009) “Affected area,” consisting of the mean activity of electrodes over the region of the tumour, which was classified by a neuro-radiologist. This corresponded to one or two of the following regions (based on EEG regions as used in Gallego-Jutglà et al., 2015): left frontal (F3); right frontal (F4); left temporal (mean of F7, T3, and T5); right temporal (mean of F8, T4, and T6); left parietal (C3); right parietal (C4); left occipital (mean of P3 and O3); and right occipital (mean of P4 and O2; see Figure 1).

Analyses were performed using IBM SPSS Statistics software (IBM, 2016). For language performance, patients’ domain z-scores were compared to pre-existing normative data from a healthy population (M = 0) by one-sample Wilcoxon signed-rank tests. With regard to slow-wave activity, i.e., delta and theta activity over the whole brain, the patient groups were compared to the control groups by using Mann-Whitney U tests. As the two tumour-specific variables could only be calculated for patients, these variables were compared to the median of the whole-brain measure of the control groups, by using one-sample Wilcoxon signed rank tests. These tests were performed after confirmation that in the control groups there were no differences between the hemispheres (N = 15, delta: Z = −1.307, p = 0.191, theta: Z = −1.367, p = 0.172; N = 9, delta: Z = −0.771, p = 0.441, theta: Z = −0.475, p = 0.635). Only the slow-wave activity measures for which patients showed significantly more delta/theta activity than healthy individuals were used in further analyses concerning language functioning. Language scores were divided into categories “impaired” (z-score ≤ −1.5 on one or more language domains) or “unimpaired” (z-score ≥ −1.5): slow-wave activity of patients with language impairment was compared to slow-wave activity of patients without language impairment with Mann-Whitney U tests. Also, the language domain scores were analysed in relation to slow-wave activity with Kendall’s tau-b correlation coefficients.³ A significance level of 0.05 was used in all tests.

Demographic and clinical characteristics of the 15 glioma patients (matched to 15 healthy individuals) and the 10 meningioma patients (matched to nine healthy individuals) can be found in Table 3. At T2, there was a drop-out of two glioma patients, due to severe side effects of adjuvant treatment, and two meningioma patients, because they were no longer motivated to participate in the study.

As expected, the low-grade glioma patients in this study have impaired language functioning at the group level before surgery. Compared to a healthy population, patients have poorer performance in almost all domains: word retrieval, phonology, semantics, and grammar. The patient group’s reading performance is better than that of a healthy population. This can be explained by the test being relatively easy as the normative data hardly include any errors and all but one patient reached the maximum score. With regard to the different language domains, impaired group-level performance in word retrieval, phonology, and semantics has previously been found in preoperative glioma patients, whereas impaired writing and comprehension of auditory input are less common (Satoer et al., 2014). Writing abilities have received little attention in patients undergoing awake glioma surgery (Van Ierschot et al., 2018), and when these abilities are evaluated preoperatively, they usually appear to be unimpaired (Santini et al., 2012; Satoer et al., 2014). The currently used writing test of Van Ierschot (2018) is presumably more sensitive to detect writing impairments than previously used writing subtests from aphasia batteries. In the year after surgery, performance on object naming recovered to a normative level, in line with earlier studies (Santini et al., 2012; Satoer et al., 2014), whereas the grammar domain, including a sentence completion test, deteriorates. This parallels a spontaneous speech analysis in which the parameter “number of incomplete sentences” significantly decreased compared to healthy controls (Satoer et al., 2018).

Before surgery, patients with a grade I meningioma have impairments in several language domains. Word-retrieval impairments have previously been found in meningioma patients preoperatively (Di Cristofori et al., 2018), which is in line with our results. Grammatical abilities and writing skills, however, have not been investigated in meningioma patients before. One year after surgery, at the group level, patients’ language performance in all investigated domains had recovered and was at the level of a healthy population, except for the writing abilities. This finding is in accordance with two other studies reporting on improvement in language skills (range: 2–12 months after surgery; Liouta et al., 2016; Di Cristofori et al., 2018). Hence, meningioma resection in the vicinity of eloquent brain areas does not seem to cause language deterioration at the group level. However, our data also show that testing of reading and writing is a useful addition to the tests already used at the spoken level.

(2) Do low-grade glioma and meningioma patients have more slow-wave activity than healthy individuals in their EEG preoperatively and 1 year postoperatively?

Glioma patients show more delta activity over the affected area, and more theta activity over the whole brain, the affected hemisphere, and the affected area before surgery. Pathologically increased slow-wave activity has already been shown to occur preoperatively in heterogeneous patient groups with different tumour types and tumour grades (Baayen et al., 2003; De Jongh et al., 2003; Oshino et al., 2007) and is now confirmed specifically for low-grade glioma patients. Nevertheless, visual EEG Examination does confirm much individual variation in the level of slow-wave activity.

Explanations for the observed slow-wave activity patterns are discussed for the delta and theta band separately. Presence of increased delta activity only over the affected area is in line with previous studies (Baayen et al., 2003; De Jongh et al., 2003; Oshino et al., 2007). Moreover, the findings, supported by Fernández-Bouzas et al. (1999), show that sources of delta activity are localised in the tumour margins. Literature on acute brain injury reports that increased localised delta activity is presumed to originate from injured tissue around the lesion, in particular, subcortical white-matter injury. Subcortical lesions appear to induce delta waves that are generated by pyramidal neurons in cortical areas overlying the lesion (partial cortical deafferentation; Ball et al., 1977; Gloor et al., 1977). Furthermore, after severe traumatic brain injury, the extent of delta activity is related to the magnitude of white-matter injury (Thatcher, 2006). Gliomas contain more glial cells than cortical grey matter and hence predominantly affect subcortical white matter (Larjavaara et al., 2007). This finding suggests that in glioma patients, white-matter injury contributes to emerging delta activity.

Increased theta activity is observed at all three spatial levels in our glioma-patient group. This findings agrees with those of Oshino et al. (2007), which assert that enhanced theta activity is present more diffusely than delta activity in preoperative brain-tumour patients. An explanation for the difference in spatial distribution between delta and theta activity is that the latter may be a result of the extent to which neural plasticity, that is, functional reorganisation of the brain, has succeeded (Duffau, 2014). In this line of reasoning, neural reorganisation was less successful in patients with residual postoperative language impairments compared to patients without postoperative language deficits. Increased theta activity may, thus, be a marker of this change in brain activity.

This current study is the first demonstrating that increased slow-wave activity is still present 1 year after glioma surgery. Moreover, the slow-wave activity patterns are identical to those before surgery (more delta activity over the affected area, and more theta activity over the whole brain, the affected hemisphere, and the affected area). De Jongh et al. (2003) reported similar patterns of delta activity preoperatively compared to 8 months postoperatively in a subgroup of patients with different types of low-grade brain tumours. Together, the findings suggest that in the long-term, brain tissue around the surgical cavity and/or residual tumour remains injured or dysfunctional. Alternatively, increased slow-wave activity 1 year after surgery may be associated with adjuvant treatment, which the majority of glioma patients in our study received around the time of testing (Nokia et al., 2012).

Unlike glioma patients, meningioma patients have no preoperative abnormal slow-wave activity at the group level, as was seen in the study conducted by Oshino et al. (2007). In their study, meningiomas were associated with less delta activity and less consistent delta source locations than intra-axial tumours (gliomas and metastatic tumours), while correcting for tumour size and surrounding oedema. The difference between tumour types is presumably due to the fact that meningiomas are not infiltrative but only cause mass effect on surrounding brain tissue, especially on subcortical tissue, and that white matter pathways tend to be displaced but intact (Campanella et al., 2014). Therefore, meningiomas cause minimal or no white-matter injury, hence, only minimal or no delta activity (Gloor et al., 1977).

(3) Is increased slow wave activity related to preoperative language functioning and 1 year postoperative language outcome?

Glioma patients with language impairment have similar delta activity, but more theta activity over the whole brain, the affected hemisphere and the affected area, compared to glioma patients without language impairment. This finding suggests that impaired and unimpaired language functioning can be further supported on the basis of theta activity level, which is a novel finding in brain tumour patients. Similar results have been reported for other patient groups. In post-stroke aphasic individuals can be discriminated from post-stroke non-aphasic individuals by increased left-hemispheric delta activity, maximally over perisylvian areas (Chapman et al., 1989). The localisation of enhanced slow-wave activity is in agreement with the current results, as the affected hemisphere is generally the left hemisphere and the affected area commonly lies around the Sylvian fissure. However, the frequency band (delta instead of theta) that Chapman et al. (1989) report is not in line with our findings. This result may be explained by the different types of brain injury (acute vs. non-acute) and by the fact that Chapman et al. (1989) did not include theta activity. Also, individuals with non-acute, neurodegenerative language impairments (e.g., primary progressive aphasia) can be differentiated from healthy, age-matched individuals by slow-wave activity patterns, more specifically, by increased delta activity and increased theta activity over left-hemispheric language-related areas (Kielar et al., 2019; Shah-Basak et al., 2019). Overall, previous studies and the current study point in the same direction: increased slow-wave activity, at least theta activity, is related to language impairments.

When considering different language domains, increased theta activity is associated with poorer word retrieval and grammatical abilities. The relation between increased theta activity and poor word retrieval, as assessed with an object-naming test, has previously been found in patients with Alzheimer’s disease (Kim et al., 2012), although not consistently (Helkala et al., 1991; Duffy et al., 1995; Van der Hiele et al., 2007) and in patients with traumatic brain injury (Thatcher et al., 1998). No significant correlations have been reported for delta activity, similar to the current study. The relation between increased theta activity and affected grammatical abilities, as assessed with action naming in sentence context and sentence completion, has not been investigated before. Previous studies did not include tests on grammar nor did they analyse grammar separately in relation to slow-wave activity (Shah-Basak et al., 2019). In general, grammatical performance is only occasionally tested in brain tumour patients, even though it contributes important information to perioperative language assessments, as it reflects communication at the sentence level (Ohlerth, 2019).

Our study showed that slow-wave activity in the theta band in low-grade glioma patients is predictive of 1 year postoperative language functioning, particularly with word retrieval (as assessed with object naming). This finding is line with earlier findings on the prognostic value of preoperative slow-wave activity on functional outcome after brain-tumour surgery (Oshino et al., 2007).

Several studies in stroke patients have also shown that increased slow-wave activity is indicative of poorer language recovery 2–24 months post-onset. Evidence has been provided for delta activity (Hensel et al., 2004), theta activity (Tikofsky et al., 1960), and delta-theta activity combined (Jabbari et al., 1979; Szelies et al., 2002). The discrepancy with our results (i.e., theta activity only) may be due to the differences in pathophysiology and the severity of the language impairment it induced: an acute stroke causing language impairments (previous studies), vs. a slow-growing tumour and its surgical resection, which caused language impairment only in some of the patients (current study). The lower rate of language impairment in low-grade glioma patients can be explained by the fact that a slowly growing tumour allows for neuroplasticity processes and reorganisation of (language) functions. Moreover, a low-grade brain tumour is generally associated with less acute brain injury and less severe language impairment than a stroke in the same location (Anderson et al., 1990).

The current results should be interpreted with caution because the sample sizes are small and there is a high level of inter-individual variability in language scores and slow-wave activity. Even though all registrations were visually inspected and checked for their quality, the slow-wave activity results may have been influenced by the use of anti-epileptic drugs at T1 and T2 and adjuvant cancer treatment at T2 (Duncan, 1987; Nokia et al., 2012). Furthermore, EEG registration comes with a few challenges. For example, EEG has relatively low spatial resolution and is sensitive to conductivity errors, as volume conduction through the skull and other tissues can blur the signal. Therefore, it is not certain that the signal registered at the scalp originates from the brain area exactly underneath. In addition, by following the standard clinical procedure, EEG was registered with 21 scalp electrodes, which limits the spatial resolution. However, an advantage of EEG in this context is that many brain-tumour patients already undergo EEG registration as part of their diagnostic examination, due to suspected epileptic seizures. In those instances, evaluation of slow-wave activity does not require much extra time. Finally, we filtered the EEG data after selecting artefact free epochs instead of on the continuous recording. we are aware that introducing and timing of filtering in the analysis can give rise to different filtering effects. It can improve the signal-to-noise-ratio but on the other hand it can also give rise to distortion which can be of concern, especially in ERP studies or analysing seizure detection on EEG recordings.