Presurgical cognitive status in patients with low‐grade glioma and epilepsy: Testing the effects of seizures, antiseizure medications, and tumor localization

Abstract Background Low‐grade gliomas (LGGs) are frequently associated with epilepsy. There are few studies addressing the impact of seizures, antiseizure medications (ASMs), and lesion localization on presurgery cognitive functioning. Methods We tested the relation between the above‐mentioned variables in a continuous series of 73 young patients (mean age 38.3 years ± 11.7) affected by LGGs and epilepsy. The anatomical areas, involved in this sample, were the left insula with surrounding cortical and subcortical areas, the right precentral gyrus/rolandic operculum, and the white matter and cortical regions beneath. Results Patients’ presurgery cognitive status was within the normal range, with borderline performance for some tasks. We tested whether lower scores were related with lesion or with epilepsy‐related factors. Multiple regression identified variables that predict test scores. The Token test score was predicted by a model (p = .0078) containing the DT2T1 MRI, corrected for seizure features. Object naming performance was predicted by a model (p = .0113) containing the localization, the DT2T1 MRI, corrected for sex, EEG, and onset. Verbal fluency score was predicted by a model (p = .0056) containing the localization and the DT2T1 MRI, corrected for AEDs and EEG. Working memory score was predicted by a model (p = .0117) containing Engel class, the DT2T1 MRI, corrected for sex. Clock drawing score was predicted by a model (p < .0001) containing the Engel class, AEDs, and EEG. TMT A score was predicted by a model (p = .0022) containing localization, corrected for EEG. TMT B‐A score was predicted by a model (p = .0373) containing localization. Voxel Lesion Symptom Mapping analyses carried out on patients’ lesion volumes confirmed that patients’ level of performance correlated with lesion‐related variables. Conclusion This preliminary study indicates that the presurgical level of performance for language tasks and for cognitive flexibility and shifting is mainly predicted by lesion‐related variables, working memory by both lesion and epilepsy‐related variables. Epilepsy clinical and instrumental characteristics predicted performance for visuospatial planning.

variables. Epilepsy clinical and instrumental characteristics predicted performance for visuospatial planning.

K E Y W O R D S
epilepsy, glioma, neuropsychology, tumor-related epilepsy, voxel-based lesion-symptom mapping INTRODUCTION There are consolidated data showing that neuropsychological alteration is a well-recognized complication of epilepsy (Lin et al., 2012).
It has also been argued that, since epilepsy involves disruption of large-scale networks, it may alter the interactions between the cognitive domains which are supported by those networks (Kellermann et al., 2016), leading to several cognitive alterations, such as memory, language, praxis, executive functions, and social cognition. Cognitive impairments differ according to the site of epileptic focus, the duration and the type of epilepsy. Cognitive alterations in tumor-related epilepsy (TRE) have not been widely investigated, especially in the subgroup represented by low-grade glioma (LGG)-related epilepsy. There is one study (Tucha et al., 2000) in which authors tested 139 patients with brain tumors of the frontal or temporal lobes, immediately after diagnosis and reported impairments of executive functions in 78% of patients, memory and attention deficits were found in 60% of patients.
Interestingly, authors reported that there was no effect of anticonvulsant drugs on cognition. The relationship between LGG and presurgical neuropsychological changes is established in the literature Racine et al., 2015;Teixidor et al., 2007;Tomasino et al., 2018;van Kessel et al., 2017). There is, however, an issue in interpreting the neurocognitive functioning of patients with LGG prior to surgical treatment, namely, the possible effect of epilepsy and antiseizure medications (ASMs) on patients' neuropsychological status. In the majority of cases (otherwise, glioma are defined incidental nonepileptic), LGG patients are affected by TRE. For instance, in a retrospective observational study with 1509 patients, the authors (Pallud et al., 2014) reported that 89.9% of them had experienced seizures at the time of tumor diagnosis.
The way TRE could affect the neurocognitive status is twofold: a direct influence, when seizures arise from the tumoral and peritumoral areas, potentially influencing the functions supported by the region.
There is a relationship between seizure incidence and LGG localization. The incidence is higher in cortical regions (56%) than in subcortical regions (15%) (Rudà et al., 2010). An indirect influence of epilepsy is expressed by functional reorganization of areas affected by seizures (Elger et al., 2004). The third way of TRE affecting the presurgical neurocognitive status is via ASMs. Since the presence of seizures affects significantly patients' quality of life, it is mandatory to reduce seizures frequency and severity and to obtain, when possible, seizure-freedom (Maschio, 2012). Indeed, inadequate or postponed epilepsy treatments can increase cognitive deterioration (Jokeit & Ebner, 1999;Maschio, 2012). ASMs could, in turn, affect the neurocognitive functioning. Often, ASMs side effects negatively affect the perception and the quality of life of patients (Maschio, 2012;Rahman et al., 2015). Among the most used drugs in TRE, there are levetiracetam, lacosamide, valproic acid, and perampanel. Mono-or polytherapy may also differentially affect patients' neuropsychological status. In the literature, however, there are contradictory evidence for an association between drug treatment and possible effects on cognitive functioning.
In the present study, we aim to evaluate the presurgical neuropsychological status of patient with TRE, and the role of localization, volume, seizure features, and drug treatment on their cognitive functioning. Our hypothesis is mainly based on the clinical observation that LGG patients result, during presurgery assessment, generally quite preserved from a cognitive perspective; thus we suggest that features related to seizures and ASMs may have no or just little effect on the cognitive state of LGG patients, affected by TRE.

Participants
In the present retrospective study, we included patients with LGG and  (Fisher et al., 2017). For statistical analysis, seizures were dichotomized in focal versus focal-to-bilateral seizures.
Preoperative EEG recordings (32-channel EB Neuro Mizar Sirius system with Galileo NT software, EB Neuro) were performed according to the 10-20 International System, within 7 days before surgery.
EEGs were scored as follows: -Normal (N): background activity with alpha or faster rhythms, no focal or diffuse slowing, no epileptic discharges.
-Slow (S): alpha or faster rhythms as background with focal or multifocal slow activity, or alpha rhythm mingled with diffuse theta-delta activity. Epileptic activity was absent.
-Epileptic (E): alpha activity in the background with faster rhythms or mixed with slower activity. Localized or diffused interictal epileptiform abnormalities (spikes, polyspikes, spike-and-wave, polyspikeand-wave complexes) were present.

MRI structural data
Data were obtained by retrospectively analyzing structural images routinely acquired presurgery. A 3-T Philips Achieva whole-body scanner was used to acquire structural data using a SENSE-Head-8 chan-

Statistical methods
For each cognitive test, we converted the Raw Score (PG) into Correct Score (PC) for age, schooling, and gender. Then each PC was converted in the correspondent Equivalent Score (PE), with a PE = 0 meaning a pathological performance. A score below or equal/above the external nonparametric tolerance limit of adjusted scores corresponds to 0 or to 4 respectively; 1, 2, and 3 are intermediate.
For some cognitive test, the normative study includes only a cutoff score. In this case, a score under the cut-off means a performance below the normal range.
Analyses were performed by using SAS software, version 9.4 (SAS, Cary, NC, USA).
We first performed a Wilcoxon-Mann-Whitney test, to test whether the medians of the neuropsychological scores (for all the tasks) differed significantly for the analyzed variables. We then performed a bivariate Spearman's correlation analysis between the epilepsy-related variables. Lastly, predictors strongly associated with test scores in univariate models and variables considered as relevant were included a multivariate logistic regression model, where the backward stepwise entry method was applied. Predictors that were significant at the .05 level were retained in the final model.

Voxel-based lesion-symptom mapping (VLSM) analysis
The VOI and behavioral data were analyzed in a voxel-based lesionsymptom mapping (VLSM) procedure https://www.nitrc.org/projects/ mricron. We used the NPM (nonparametric mapping) software. We set the Brunner Munzel (BM) test, 1000 permutations, only testing voxels damaged in at least 10% individual for each test (10% of patients within that hemisphere for test performed to the LH or RH patients [see Table 2] or 10% of all patients for working memory and short-term memory). Each patient's accuracy score on the cognitive task was used as continuous behavioral variable. The critical z-value considered as our BM map was p < .05 value. Any value in the power map and the BM map exceeding this critical z-value was considered significant.

Study population
A consecutive series of 73 low-grade glioma (LGG) patients (48 male, 25 female, mean age 38.3 years ± 11.7; mean education was 13 ± 3.8) entered in the study. All patients had TRE and were pharmacologically treated. Mean presurgical duration of epilepsy was 5.8 ± 5 months. Table 1 shows information about lesion localization, size, onset of the disease, and Table 2 shows patients' seizure frequency and pharmacological therapy.

MRI structural results
The lesion overlay showed that the maximum overlap occurred in the left insula and surrounding cortical and subcortical area, and the right precentral gyrus/rolandic operculum and the white matter and cortical region beneath (see Supplementary Table S1 and Figure 1).

Neuropsychological results
In general, the patients' presurgery neuropsychological status was within the normal range: at Raven matrices, short-term memory and working memory, the whole group of patients succeeded. Tasks administered to the RH and the LH groups showed that, for the RH group, all the patients were normal in almost all the measured domains (see Table 3) except for a low number (3/17) of pathological patients at processing speed. For tests allowing the use of equivalent scores (namely short-term memory, working memory, verbal fluency, verbal comprehension, TMT, and constructional praxis), we used them to identify borderline performance (see Figure 2). A score lying below or equal/above the external nonparametric tolerance limit of adjusted scores corresponds to 0 or to 4 respectively; 1, 2, and 3 are intermediate (Capitani & Laiacona, 1988) (see Table 3). This suggests that for some tasks such as verbal fluency, despite patients were still within the normal range, their performance was borderline (ES = 1). We performed a bivariate Spearman's correlation analysis between the epilepsy-related variables.

F I G U R E 2
Patients' performance expressed as equivalent scores 1-4. A score lying below or equal/above the external nonparametric tolerance limit of adjusted scores corresponds to 0 or to 4 respectively; 1 is borderline, 2 and 3 are intermediate We then performed a multiple regression (backward technique).
The model containing the variable codifying Engel class, AEDs, and EEG significantly (p < .0001) predicts the clock drawing score.
The model containing the variable codifying the localization (pre-/postcentral vs. temporo/insular area), corrected for EEG, significantly predicts TMT-A score (p = .0022). In particular, a lesion localization in temporo/insular area corresponds to an increase in score of 12.9 (p = .0013, Parameter estimate = 12.89).
The model containing the variable codifying the localization (pre-/postcentral vs. temporo/insular area), significantly predicts TMT B-A score (p = .0373). In particular a lesion localization in temporo/insular area corresponds to an increase in the score of 28.5 (p = .0373, Parameter estimate = 28.500).

Neuroanatomical results
Results of the VLSM analyses (see Figure 3 and Note: The number of damaged voxels (and the % of the total no. of voxels) for each region as reported in the brain atlas of grey (AAL) and white matter (JHU and NatBrainLab) are reported, along with the center of mass (Max X, Y, and Z MNI coordinates).
area), the putamen (66% of the area), the external capsule (56% of the area), the olfactory area (40% of the area), the pallidum (32% of the area), the anterior corona radiate (39% of the area), the pars orbitalis of the inferior frontal gyrus (31% of the area), the caudate (36% of the area), and the uncinate fasciculus (49% of the area).
All the other tasks did not survived correction for multiple comparisons.

DISCUSSION
We studied a consecutive series of 73 patients with LGG involving the left and the right hemispheres. All patients had seizures at the time of onset and were on drug treatment with ASMs.
Our main result is that we found that performance in tasks related to language (Token test, object naming, and verbal fluency), cognitive flexibility and shifting (TMT) was predicted by lesion-related variables.
Epilepsy-related variables predicted performance for one task only, namely visuospatial planning performance.

The little relation between ASMs (and epilepsy-related variables)
and cognitive impairment result may depend on three main aspects.
First, the majority of patients were treated with only one medication. This might limit the occurrence of cognitive adverse events and drug-to-drug pharmacodynamic and pharmacokinetic interactions that may result in side effects potentiation and increased drug blood levels (van Breemen et al., 2009). Nowadays, in treating LGG-related epilepsy, a monotherapy approach is preferred. For example, in a group of 140 patients (33 LGG), (van Breemen et al., 2009) a greater use of ASM monotherapies in patients with LGG is reported (56% LGG vs. 36.7% high-grade Glioma [HGG]). These data suggest that monotherapy approach could preserve patients from the onset of cognitive deficits that may be related to ASM.
Second, in our sample, the most represented ASM was levetiracetam (LEV), at mean dose of 1500 mg/die, which does not affect significantly cognitive functions (Lamberty et al., 2000;Zhou et al., 2008). The wide use of LEV may represent a bias in our sample; however, LEV is nowadays one of the most prescribed ASMs for TRE. LEV and other newer ASMs (i.e., lacosamide, perampanel, and brand new brivaracetam) are reported by guidelines and expert opinions as the best choice to treat TRE, for their favorable pharmacokinetic profile and lack/limited cognitive side effects (Maschio et al., 2019). By contrast, old generation ASMs, especially enzyme-inducing ASMs, have side effects exerting also on cognitive functioning and on patients' mood as they may generate depression and irritability (Maschio, 2012;Maschio et al., 2019).
Lastly, in our sample, mean duration of epilepsy was 5.8 months, thus the role of recurrent seizures in generating possible cognitive impairment appears less important comparing with long-term epilepsies. A recent study (Gavrilovic et al., 2019)  For example, mood cognition and quality of life were examined in a study (Rahman et al., 2015) involving a cohort of 81 patients affected by gliomas (both LGG and HGG), medulloblastomas and meningiomas with 55 of them presenting also TRE. It was reported that cognitive impairments occurred in 50.6% of their subjects, but this happened regardless the presence of epilepsy and the use of ASMs. It is worthy to mention that, unlike our population, many patients were under polytherapy and were taking carbamazepine and valproate. It is also worthy to mention that cognitive status was measured using the MoCA and frontal assessment battery tests. We complement these results, by adding extended information about the patients' neuropsychological status.
In another study (Maschio et al., 2014), investigating the burden of epilepsy in patients with brain tumors, quality of life was significantly affected by presence, type, and duration of therapy with ASMs. In particular, patients perceived a negative effect on cognition and social function. Interestingly, this occurred independently of the ASM used.
Moreover, this subjective perception was stronger in patients receiving ASMs for periods longer than 6 months. These data came all from subjective scale evaluating quality of life, which were collected at the first neurological visit. No statistically significant difference has been observed regarding Mini Mental State Examination scores. Thus, neither the tumor nor epilepsy and its treatments influenced the neurological functions of the subjects examined and ASMs seemed to affect patients' cognitive status more from a subjective point of view (Khan & Anatya, 2013;Maschio et al., 2014). In our sample, short duration of epilepsy, prevalent antiseizure monotherapy, and first diagnosis of LGGs, with all the expectations related to surgery, which represents, in most of cases, the prominent therapeutic act, may contribute to reduce the perception of illness burden.
In a further study (Derks et al., 2019)

Neuropsychological results
From a neuropsychological point of view, patients perform within the normal range in most tests.
The normal cognitive performance of our patients supports the idea that a monotherapy treatment is preferable to a polytherapy. This avoids the toxicological effect of the use of several drugs at the same time (Klein et al., 2003). Not only new ASMs cause fewer side effects on cognitive functioning than old ASMs (Klein et al., 2003). This is also confirmed in our work where patients' cognitive performance is mostly within the normal range. The most of our sample (90.5%) have a monotherapy drug treatment. 54/73 (73.9%) patients assumed LEV. As reported in literature (Vecht & Wilms, 2010), LEV has a good tolerability, reduces 50% or more seizures in 2/3 of patients, and has significant fewer cognitive side effects. LEV is a second-line ASM and presents higher efficacy in patients with seizure (van Breemen et al., 2009).

Neuroanatomical results
In our LH-group VLSM maps revealed an involvement of the precentral gyrus, frontal (superior and middle) gyrus, inferior frontal, insula, anterior cingulate, caudate, putamen/globus pallidus, and left-thalamus in verbal fluency. This is a classical neuropsychological test of language production where subjects have to generate and articulate words in response to a phonological cue. Neurocognitive models define an important role of the left frontal executive regions in phonological fluency (Mummery et al., 1996). Results are similar to other studies about verbal fluency, showing a role of the left insula, putamen, superior temporal pole, and external capsule (Pisoni et al., 2019), and the left putamen, caudate nucleus and pallidum, and temporal region (Chouiter et al., 2016).
In our VLSM analyses, we found an involvement of the insula, putamen, and frontal structures, consistently with previous functional neuroimaging literature (Ivanova et al., 2018;Wager & Smith, 2003). Other authors (Grahn et al., 2009) and (Ivanova et al., 2018) have found activation in WM task also in caudate, as we did in our sample in addition to frontal areas.
As far as the Token test is concerned, in a VLSM study (Pisoni et al., 2019) authors found task impairments correlated with damaged voxel in posterior part of the left superior and middle temporal gyri as well as with the angular and supramarginal gyri. Other authors (Papagno & Cecchetto, 2019) argued that as the Token test involves short-term memory component, significant voxels are found in temporo-parietal areas (Pisoni et al., 2019). Authors identified voxel damage correlating to impaired performance at the Token test in the middle and superior temporal gyrus, the supramarginal gyrus, angular, and in subcortical areas. The large network involved in language comprehension includes also subcortical structures such as basal ganglia, especially the caudate and the thalamus, as we found in our VLSM analysis.

COMPETING INTEREST
None of the authors has any conflict of interest to disclose.

ACKNOWLEDGMENTS
The medical imaging staff is kindly acknowledged. B.T. is supported by the Italian Ministry of Health (RicercaCorrente).

AUTHOR CONTRIBUTIONS
G.P. T.I., and B.T. designed the research; I.G., G.P., T.I., and M.S. performed the research; F.P. analyzed the data; I.G., B.T., and G.P. wrote the paper; all authors edited the paper; I.G., B.T., G.P., T.I., and M.S. collected the data; all authors revised the final version of the manuscript.

DATA AVAILABILITY STATEMENT
Data will be shared upon request by contacting the corresponding author.