Introduction

Pharmacoresistant mesial temporal lobe epilepsies (mTLE) represent the most frequent indication for epilepsy surgery. Given the essential role of the limbic system for declarative memory functions [1], the surgical treatment for mTLE particularly bears the risk of memory decline. According to a recent meta-analysis [2], the risk of significant deteriorations in verbal memory is 44 % for epilepsy surgeries within the left temporal lobe and 20 % for right temporal lobe resections, whereas significant declines in nonverbal memory functions are to be expected after 21–23 % of temporal lobe surgeries.

Three major factors have been identified as determinants of the cognitive and mnestic outcome after surgical treatment of mTLE [3–5]: (1) the functional integrity of the resected tissues (also termed “functional adequacy”), (2) cognitive reserve capacities, and (3) postsurgical seizure control.

Postoperative seizure freedom can lead to a functional recovery of brain regions that were secondarily affected by the active epilepsy [3, 6, 7].

Cognitive reserve capacities refer to the efficiency and plasticity of the brain to counteract iatrogenic negative effects of epilepsy surgery [8]. Higher reserve capacities are reflected by better presurgical functioning as well as by higher intelligence and education. Finally, a younger age and female sex are associated with greater functional plasticity [9–11]. In this regard age at onset and duration of epilepsy need to be considered, reflecting time windows and critical phases for reorganization processes [8].

In this context the functional integrity of the structures which are going to be resected has a major impact on the cognitive outcome of epilepsy surgery [12]. A postoperative decline is to be expected when the resection area still subserves cognitive functions. The cognitive deterioration after surgical removal of non-lesional tissue in MRI- and histopathology-negative patients with TLE may serve as proof of principle [13]. Consequently, the severity and extent of the underlying pathology can represent an estimate of the functional integrity of the to-be-resected structures. One approach to quantify pathology in mesial temporal lobe epilepsy is the analysis of neuropathological specimens, i.e. semi-quantitative measurements of hippocampal cell loss. In two studies by Hermann et al. verbal (and in the earlier study also nonverbal) memory decline after left temporal lobe epilepsy surgery was greater in patients with absent or mild (Wyler I or no HS) versus severe (Wyler II–IV) hippocampal pathology [14, 15]. This differential outcome was not observed in patients with right temporal lobe resections. Similar findings were reported by Rausch and Babb [16] and by Sass et al. [17] who performed a median split on quantified neuron densities within the resected hippocampus to distinguish patients with severe versus mild to moderate hippocampal pathology. Both studies focused on verbal memory and again solely the left hippocampal integrity was relevant for the postsurgical memory outcome. Two other studies [18, 19] directly correlated neuronal cell densities of hippocampal specimens with postoperative memory changes, indicating that greater decline in verbal memory is associated with higher neuron densities in CA1 (cornu ammonis). The study by Zaidel et al. [19] in addition also found significant correlations between verbal memory changes and cell loss in CA4 and the dentate gyrus, while the subfields CA2 and CA3 had not been investigated.

Despite different methodological approaches, all cited studies emphasize the importance of the hippocampal integrity for verbal memory change after left temporal lobe surgery. The role of the structural integrity of the right hippocampus for postsurgical memory change remains unclear.

Previous studies often had small and heterogeneous samples, partially with mixed samples of mesial and lateral TLE, and they used memory tests with different sensitivities to mesial temporal lobe pathologies [20, 21]. In the current study, we examined postoperative memory change as a function of neuronal cell densities within hippocampal subfields in a large sample of patients with solely mesial temporal lobe epilepsy (mTLE), employing verbal and figural memory tests with proven sensitivity to mesiotemporal pathology. Since our previous study in the same sample had revealed that the overall integrity (rather than specific subfields) of the resected hippocampus is an important determinant of preoperative memory performance [22], we again analyzed the mean neuronal cell loss across all subfields.

According to the functional adequacy theory, we expected a positive correlation between the degree of postsurgical memory declines and neuronal cell densities. In addition, moderating effects of the side of surgery and the individual pattern of hemispheric dominance for language and verbal memory were assumed [22, 23].

Methods

Patients

The analyses were based on 104 consecutive patients who underwent surgical treatment of unilateral mesial temporal lobe epilepsy at the neurosurgery department of the University of Bonn [22]. The comprehensive presurgical evaluation with non-invasive and/or invasive procedures revealed seizure onset in the mesial temporal lobe in all patients [24]. The resection of the hippocampus was clinically indicated in every case. All procedures were conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the University of Bonn Medical Center. Informed written consent was obtained from all patients.

Inclusion criteria comprised a chronological age of at least 16 years, complete data on neuronal cell densities for all hippocampal subfields, and complete pre- and postsurgical neuropsychological assessments on verbal and figural memory.

Side of surgery was balanced with 53 left (51 %) and 51 right (49 %) temporal lobe resections. The most frequent type of surgery was selective amygdalohippocampectomy (80 %), followed by standard two-third resection of the temporal lobe (8 %), resection of the temporal pole plus amygdalohippocampectomy (7 %), and extended lesionectomy plus amygdalohippocampectomy (6 %).

Demographic and clinical characteristics of the study sample are given in Table 1. More detailed information on the antiepileptic treatment at the time of the pre- and postsurgical assessment is provided in Supplementary Table 1. There were no significant differences in demographic and clinical characteristics between patients with left versus right temporal lobe surgeries.

Table 1 Demographic and clinical characteristics and neuronal cell densities of the resected hippocampus
Full size table

The sample was composed of 54 female (52 %) and 50 male patients (48 %) with a mean age of 38.2 ± 12.8 (standard deviation). The average age at onset of epilepsy was 15.5 ± 12.5 years, the mean duration of epilepsy 22.7 ± 13.9 years. Prior to surgery, patients suffered from 8.9 ± 11.4 seizures per month. At the time of the postsurgical neuropsychological assessment, i.e. 12.9 ± 3.7 months after surgery, 65 % had been completely seizure free (Engel Ia/ILAE Ia; Table 1). This was accompanied by a significantly reduced mean number of concurrent AEDs (t = 3.818, p < 0.001; Table 1).

As part of the presurgical evaluation, 47 % of the patients underwent assessment of language dominance (32 % language fMRI, 17 % Wada; 2 % both). In patients without determination of language dominance, atypical hand dominance in left mTLE (4 %) or a neuropsychological suppression pattern (i.e. intraindividual profile with significantly better verbal than nonverbal memory performance despite left mesial temporal lobe epilepsy) (4 %) served as markers of atypical dominance regarding language or material-specific memory functions [25–27]. Despite typical language dominance as revealed by fMRI or Wada, the intraindividual material-specific memory profile of 4 patients with left mTLE (4 %) indicated a selective memory transfer [28, 29]. All in all, 16 patients (30 %) of those with left mesial temporal lobe epilepsy presented with markers of an atypical dominance for language or material-specific memory functions.

Pre- and postoperative memory assessment

Neuropsychological examinations of verbal and nonverbal memory functions were conducted as part of the presurgical evaluation and again around 1 year after surgery. To minimize practice effects parallel versions of the memory tests were employed at the postoperative assessment.

Verbal memory performance was assessed using the Verbaler Lern- und Merkfähigkeitstest (VLMT) [30], a modified German version of the Rey Auditory Verbal Learning Test—RAVLT [31]. The VLMT is the most frequently applied verbal learning and memory test in German epilepsy centers [32]. The test requires serial learning and immediate recall of 15 words in five consecutive learning trials, free recall after distraction as well as free recall and recognition of the target words after a 30-min delay. Analyses were based on learning performance (total number of words learned in five trials), delayed free recall, and recognition performance (corrected for false-positive answers). The test has been shown to be sensitive to left temporal lobe dysfunction, left mesiotemporal pathology and left-sided temporal lobe surgery [33–36].

Nonverbal figural memory was assessed by the revised version of the Diagnosticum für Cerebralschädigung (DCS-R) [37]. The test requires learning and reconstruction of 9 abstract designs in 5 consecutive trials. After a retention interval of 30 min, recognition memory is tested. The considered measures were the number of correctly learned designs over 5 learning trials and recognition performance (corrected for false-positive answers). The test has been shown to be sensitive to right temporal lobe dysfunction, right mesiotemporal pathology, and also to right-sided temporal lobe surgery [36–39].

Postoperative changes were determined by the difference between pre- and postsurgical scores (postoperative raw score minus preoperative raw score, i.e. positive values indicate gains, negative values indicate losses, and a value of zero indicates no change). To evaluate significant intraindividual changes, both memory tests provide practice-corrected reliable change indices (RCI) based on 100–142 reassessed healthy controls.

At the time of the neuropsychological assessments, the examiner was not aware of the neuropathological results.

Neuropathological examination

Hippocampal specimens were microscopically examined at the Department of Neuropathology, University Hospital of Erlangen. Neuropathological investigators were blinded to pre- and postoperative neuropsychological results.

En bloc specimens of the hippocampus were dissected into 5 mm thick slices along the anterior-posterior axis. Only tissue from the mid-hippocampal body was selected for this study to warrant unequivocal recognition of all subfields of the hippocampal pyramidal cell layer as well as the granule cell layer. The tissue was fixed overnight in 4 % formalin and routinely processed into liquid paraffin. Four-μm sections were stained with hematoxylin and eosin (HE) for histopathological inspection. Hippocampal pyramidal neurons and granule cells of the DG were specifically detected using immunohistochemistry for the neuronal core antigen NeuN (A60, Chemicon, Temecula, USA, dilution 1:1000, pre-treated with microwave) and an automated staining apparatus using the streptavidin–biotin method (Ventana; Strasbourg, France) and 3,3′-diaminobenzidine as chromogen as well as hematoxylin counterstaining.

One representative hippocampal specimen was chosen from each subject for semi-quantitative neuronal cell counts. Semi-quantitative measurements of neuronal cell numbers were performed with a microcomputer imaging system (ColorView II CCD camera, AnalySIS imaging software, Stuttgart, Germany) equipped to a BX51 microscope (Olympus, Japan). Only NeuN-immunoreactive neuronal cell bodies were tagged on the computer screen and counted separately within hippocampal sectors CA1, CA2, CA3 and CA4 in four randomly placed visual fields at 20× objective magnification (Fig. 1). We separately examined granule cells of the external (DGext) and internal (DGint) limb at 40× objective magnification in ten randomly placed visual fields (Fig. 1). The hippocampal subfield specific quantification method was previously described to match data obtained from stereological cell counts and to be suitable for clinico-pathological correlations when using formalin-fixed and paraffin-embedded surgical human tissue specimens [40–42]. Differentiation of human hippocampal subfields followed the first international classification system for hippocampal sclerosis [42]. Subfield specific analyses were based on absolute cell densities. However, to determine the overall integrity of the hippocampus, neuronal cell counts (neurons/mm2) were in the first instance transformed into standard-scores (mean = 100, standard deviation = 10) according to control values obtained from autopsy brains. Finally, since principal component analysis of segmental neuronal cell loss within the hippocampus had revealed a single factor structure [22], the mean standardized neuronal cell density across all hippocampal subfields was calculated. This index was termed “hippocampal integrity index” (HCI).

Fig. 1
figure 1

Surgical specimen of a a non-pathological human hippocampus (control) as compared with b severe hippocampal sclerosis (HC). The figure shows the different anatomic subregions of the hippocampus including cornu ammonis (CA) sectors CA1–CA4 as well as the dentate gyrus (DG). a NeuN immunohistochemistry of the hippocampus highlighting the subiculum (SUB), the sectors of the Cornu ammonis (CA1–CA4) and the dentate gyrus (DG) without evidence for segmental cell loss. Inset in a higher magnification of the dentate gyrus. Regular package density of the granule cell layer without signs of granule cell loss or dispersion. c–f Higher magnification (of a) of the Cornu ammonis sectors with regular pyramidal cell densities. b NeuN immunohistochemistry evidencing neuronal cell loss in sectors CA1–CA4. Inset in b higher magnification of the dentate gyrus with areas of severe granule cell depletion. g–k Higher magnification (of b) of the cornu ammonis sectors with severe pyramidal cell loss in all subfields (CA1, CA3 and CA4 more severe than CA2). a, c, d specimen with no hippocampal sclerosis. b, g–k specimen with hippocampal sclerosis (ILAE HS Type 1). Of note a and b are of same magnification, indicating also the ‘shrinkage’ of the hippocampal body in a specimen with hippocampal sclerosis. Scale bar in a 1 mm, applies also for b. Scale bar inset in a 100 µm, applies also for inset in b. Scale bar in c 200 µm, applies also for d–k

Full size image

Statistical analyses

Repeated measures multivariate analyses of variances (MANOVA) were used to analyze pre- to postoperative memory change. Correlation analyses (Pearson) were applied to investigate the relationship between neuropsychological and neuropathological data.

Results

Neuronal cell densities

The neuronal cell densities of the resected hippocampal specimens are presented in Table 1. The most affected subfield was CA1 with a mean neuronal cell loss of 72 % followed by CA4 (67 %), CA3 (58 %), DGint (58 %), DGext (51 %) and finally CA2 with the least cell loss (42 %). The pattern of neuronal cell loss was not different for left and right hippocampal specimens.

Pre- to postoperative change of memory performance

The pre- and postsurgical mean memory performance of patients with left versus right mTLE is illustrated in Fig. 2. A repeated measures MANOVA with the between-subject factors “side of surgery” and “postsurgical seizure freedom” revealed a significant impact of surgery on memory performance (F = 3.31, p = 0.003) and a related interaction effect of the side of surgery (F = 2.72, p = 0.013), but not of seizure control (F = 0.96, p = 0.468). In the total sample, all memory parameters showed a postoperative decline. When considering the side of surgery, patients with left hemispheric resections showed greater deteriorations in verbal learning (F = 12.53, p = 0.001) and recognition (F = 8.60, p = 0.004) than those with right-sided operations. Regarding verbal delayed free recall a trend into the same direction was observed (F = 3.91, p = 0.051). Changes in figural memory were not dependent on the side of surgery.

Fig. 2
figure 2

Standardized pre- and postsurgical memory performance of patients who underwent left- versus right-sided surgery for mesial temporal lobe epilepsy (mTLE). Memory parameters were standardized according to a co-normalization sample of 488 healthy volunteers (mean = 100, standard deviation = 10), applying a correction for age. Higher values reflect better performance. Values <90 indicate below average performance compared to the healthy normative sample. (*) p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001  

Full size image

Significant changes of memory performance on an individual level are summarized in Table 2. According to RCIs, 42–62 % of the patients deteriorated in verbal memory functions after left-sided surgery versus 9–13 % after right temporal resections. The respective numbers for figural memory were 9–11 % versus 10–20 % after surgical treatment of left versus right mTLE. However, regarding the decline of figural learning performance floor effects need to considered in 26 % of patients with left mTLE and 45 % of patients with right mTLE. In those patients significant postsurgical deteriorations in figural learning could no longer be demonstrated, because of the already severely impaired presurgical performance.

Table 2 Frequencies of significant individual changes after temporal lobe surgery according to reliable change indices
Full size table

Postsurgical memory change as a function of neuronal cell densities

Correlation analyses indicated significant relationships between neuronal cell counts of the left hippocampus and the postsurgical decline of all three verbal memory parameters. Higher cell densities were associated with greater deteriorations. Correlation coefficients ranged from r = −0.27 to r = −0.39 (Table 3). Among the hippocampal subfields only CA3 and DGext were unrelated to verbal memory change. Solely one significant correlation was found between neuronal cell loss of CA4 and figural (recognition) memory (r = −0.31, p < 0.05).

Table 3 Correlations between neuronal cell densities of the ipsilateral hippocampus and metric pre- to postsurgical memory change (postsurgical minus presurgical test performance) after left (N = 53) versus right (N = 51) temporal lobe surgery
Full size table

In the subgroup of patients with left mTLE and no evidence of atypical hemispheric dominance (N = 37) markedly higher correlations between hippocampal cell loss and change in verbal memory indices were observed, ranging from r = −0.34 to r = −0.60 (Table 4). In that subgroup, changes in verbal learning and delayed free recall were correlated with cell counts of all hippocampal segments. The hippocampal integrity index (HCI) explained up to 34 % of the observed variance in verbal memory change.

Table 4 Correlations between neuronal cell densities of the left hippocampus and pre- to postsurgical memory performance (postsurgical minus presurgical test performance) in patients without evidence of atypical hemispheric dominance who had undergone left temporal lobe surgery (N = 37)
Full size table

Segmental cell loss within the right hippocampus showed no significant correlations with verbal memory change. Regarding figural memory, only one significant correlation could be discerned: higher cell counts within CA3 were associated with greater declines in figural recognition.

Discussion

In this study, we investigated the role of hippocampal integrity—as determined by hippocampal cell counts—for postoperative memory outcome in the up to now largest sample of 104 patients with solely mTLE.

When considering the total sample, postoperative memory decline was indicated in all main parameters of the employed memory tests. In line with the literature [2, 4], group and individual level analyses showed a differential negative effect of left-sided surgery on verbal memory. Changes in figural memory appeared to be independent from the side of surgery. This may be partly due to floor effects since nearly half of the patients with right mTLE already displayed severe presurgical impairments that could not become significantly worse.

Analyses revealed no impact of postsurgical seizure freedom (achieved in 65 % of the patients) on memory changes.

The number of patients with significant intraindividual memory decline was within the range of the already cited meta-analysis [2] and in line with our monocentric outcomes in more than 700 patients [43]: verbal memory declined in 42–62 % versus 9–13 % of the patients who had undergone left versus right temporal lobe surgery. Deteriorations in figural memory were observed in 9–11 % versus 10–20 % of the patients after surgical treatment of left versus right mTLE.

In accordance with the functional adequacy theory, greater declines in memory functions were associated with higher neuronal densities within the resected hippocampus. However, this was primarily the case for verbal memory changes after left temporal lobe surgery, and especially when left hemisphere dominance for language and memory was suggested [22, 23].

As for figural memory, the present analysis showed that only the cell densities of one subfield (CA3) of the right hippocampus correlated with the postoperative change, and this was solely observed for figural recognition memory. Again a higher cell density was associated with greater postsurgical declines. However, this single correlation should not be overinterpreted, since all other subfields were unrelated to memory change. Restricted variance due to floor effects in figural memory in patients with right mTLE could have contributed to the lack of a systematic relationship. Presurgical memory profiles, however, were consistent with the material-specific memory model (Fig. 2), i.e. worse figural and better verbal memory performance in patients with right compared to left mTLE [22]. The finding that only verbal memory decline could be predicted by hippocampal cell counts confirms the findings by Baxendale et al. [18].

In summary, the current study emphasizes the role of the structural and functional integrity of the left hippocampus not only for preoperative functioning but also for the degree of verbal memory decline after temporal lobe surgery. Thus, from a diagnostic point of view, presurgical verbal memory performance may be taken as a marker of left hippocampal integrity in left mTLE (in case of typical hemispheric dominance for material-specific memory functions) [22]. Since hippocampal cell counts cannot be determined before epilepsy surgery, presurgical biomarkers of the structural and functional integrity of the hippocampus would be appreciated that would advance individual risk–benefit evaluations before elective epilepsy surgery (e.g. high-field MRI [44, 45], volumetric and FLAIR analysis [46], spectroscopy [47], relaxometry [48], diffusion tensor imaging [49], memory fMRI [50], and EEG-analyses [51, 52]).

The findings of the current study also underscore that there is still the need to find a more precise neuropsychological marker of right hippocampal integrity [53–56].

Future studies would be appreciated which address the question whether a more thorough presurgical determination of the structural and functional integrity of both hippocampi could be provided by a multimethodological approach with additional use of structural and functional imaging techniques and physiological measures.