Within <1 mm radius from the atlas-planned anteroposterior and mediolateral coordinate. The success rate was evaluated based on the location of the lower tip of bipolar electrode. Abbreviations: DG, dentate gyrus**.** Statistical significance: \* *<sup>p</sup>* < 0.05; \*\*\* *<sup>p</sup>* < 0.001 compared to MRI cohort atlas-based without MRI guidance (ለ<sup>2</sup> test).

*MRI cohort.* Histological analysis indicated that the tip of the anterior intracortical electrode was within the cerebral cortex in 76% (32/42) of the cases. In the remaining 24% (10/42), the tip was located in either the external capsule or the corpus callosum. The histological AP and ML electrode tip locations were within 0.5 mm of the targeted atlasbased coordinates in 10% (4/42) and 86% (36/42) of the rats, respectively (Figure 2A,B). The histological DV location of the cortical electrode tip was in the planned target (i.e., layer V) in 31% (13/42) of the animals. In the remaining 69% (29/42), the tip was located in either layer VI, the external capsule, or the corpus callosum (Figure 2C and summary Table 1).

#### 3.1.2. Anteroposterior Location of the Anterior Intracortical Electrode Tips

Based on estimation of the progression of the cortical lesion [7,27,32], the anterior intracortical electrode was aimed to the atlas-based AP level 1.72 mm from the bregma (primary somatosensory cortex, S1) to ensure proper EEG recording in the lesion vicinity during the 7th post-TBI month (Figure 1C).

*EEG cohort.* Histological examination showed neurodegeneration along the electrode path and around the electrode tip, often accompanied by iron deposits (Figure 3A,D,F). Occasionally, the electrode-associated lesion merged with the TBI-induced lesion cavity. The unfolded cortical map indicated that 100% of the electrode tracks were in S1. Further, 83% (47/57) of the electrode tips (at least one of the tips) were found in S1 (Figure 4A,B). The remaining 17% (10/57) of the lower electrode tips were located in either the external capsule or the corpus callosum (Figures 2C and 4A,B). The upper tip of the bipolar electrode (1.0 mm from the ventral tip), however, recorded in the targeted S1 cortex. In most of the rats, the histological AP electrode location was anterior to the targeted atlasdefined coordinate (−1.7 mm) (Figure 2A,D), the average deviation being 1.21 ± 0.07 mm (range: 0–2.56 mm, median: 1.24 mm). The rostral shift was comparable between the shamoperated experimental controls and TBI rats (sham 1.16 ± 0.15 mm vs. TBI 1.22 ± 0.09 mm; *p* > 0.05) (Figure 2A,D).

**Figure 3.** (**A**–**F**) Electrode tracts. Histological images from the coronal thionine-stained sections of 6 rats, showing the tracts of the bipolar intracortical electrodes and location of the lower electrode tip (filled arrowhead). Roman numerals indicate the cortical layers. In panels (**A**,**B**) the electrode tip is located in layer V of the perilesional cortex. Note the electrode track-related lesion on the surface of the brain in panel (**A**) (open filled arrow). In panel (**A**), the electrode tip is within 500 μm from the edge of the TBI-induced lesion cavity (asterisk). In panel (**C**), the electrode tip is in the external capsule (ec). In panel (**D**) the electrode tip is within the cortical lesion, close to the angular bundle. Open arrow points to the electrode path associated neurodegeneration. In panel (**E**), the electrode tip is close to the edge of the lesion cavity (asterisk). In panel (**F**), the electrode tip is within the angular bundle (closed arrowhead). The open arrowhead points to the location of the upper electrode in layer IV (open arrow). The dark staining indicates iron deposits (arrowheads) adjacent to the electrode path. Scale bar = 500 μm.

**Figure 4.** Location of the lower tip of the anterior and posterior intracortical electrodes on atlas plates and unfolded cortical maps. (**A**) In the EEG cohort (upper panel), the dorsoventral (DV) location of at least 1 of the tips of all anterior bipolar electrodes (atlas plate: bregma −1.4 mm) was within the primary somatosensory cortex (S1) and that of the posterior electrode (lower panel; atlas plate: bregma −6.8 mm) was within the visual cortex. Each dot represents 1 bipolar electrode. (**B**) An unfolded map (UFM) showing the location of electrode tracks in the EEG cohort as seen from the surface of the brain. The intersection of the electrode path with cortical layer V was used as reference. The UFMs confirmed the location of the anterior electrode paths in the S1 and posterior electrode paths in the visual cortex. (**C**) Atlas plate showing the DV locations of the anterior (upper panel) and posterior (lower panel) intracortical electrodes in the MRI cohort. As in the EEG cohort, the anterior electrode was in S1 and the posterior electrode was in the visual cortex. (**D**) A UFM showing the location of electrode tracks in the MRI cohort as seen from the surface of the brain. All electrode tracks were within S1 or the visual cortex. Note that in the MRI cohort, we used the 5-month in vivo MRI to adjust the electrode coordinates to target the perilesional cortex and to avoid lesion cavities, underlying brain areas, or ventricles. As expected, this resulted in a more heterogeneous distribution of electrode paths than in the EEG cohort with atlas-based fixed coordinates. Atlas plates and UFMs were generated using the Paxinos rat brain atlas (6th edition).

*MRI cohort.* Electrode-associated cortical lesions were rare in the MRI cohort (Figure 3A,D). As summarized in the cortical unfolded map, 100% of the electrode tracks (100%) and 76% (32/42) of the electrode tips were within S1 (Figure 4C,D). In the remaining cases, the lower tip was located within the external capsule or the corpus callosum (Figure 4C,D). Still, the upper tip of the bipolar electrode (0.5 mm from ventral tip) should have recorded in the targeted S1 cortex. As in the EEG cohort, the histological electrode location was anterior to the targeted MRI-guided AP coordinate, the average being 1.18 ± 0.08 mm (range: 0.2–2.16 mm, median: 1.19 mm) (Figure 2A,D). The rostral shift was comparable between the sham-operated controls and TBI groups (sham 0.99 ± 0.14 mm vs. TBI 1.25 ± 0.09 mm; *p* > 0.05) (Figure 2A,D).

#### 3.1.3. Mediolateral Location of the Anterior Intracortical Electrode Tips

*EEG cohort.* Histological assessment revealed that the tip was lateral in 93% (53/57) of the cases and medial to the atlas-based coordinates (4 mm from midline) in 7% (4/57) of the cases (Figure 2B). The mean distance between the histological and MRI-guided ML coordinate was 0.58 ± 0.05 mm (range: 0–1.50 mm, median: 0.5 mm) (Figure 2D). The distance was comparable between sham-operated controls (0.68 ± 0.09 mm) and TBI (0.55 ± 0.06 mm) rats (*p* > 0.05) (Figure 2D).

*MRI cohort.* Histological assessment revealed that the tip was lateral to the MRIbased coordinates in 41% (17/42) of the cases and medial in 59% (25/42) of the cases (Figure 2B). The mean distance between the histological and MRI-guided ML coordinate was 0.44 ± 0.04 mm (range: 0–1.0 mm, median: 0.4 mm) (Figure 2D). The distance was comparable between the sham-operated controls (0.48 ± 0.09 mm) and TBI rats (0.42 ± 0.05 mm; *p* > 0.05) (Figure 2D).

#### 3.1.4. Dorsoventral Location of the Anterior Intracortical Electrode Tips

Cortical layer V was set as the DV target for the lower tip of the bipolar intracerebral electrode in both cohorts.

*EEG cohort.* Histological analysis revealed that 83% (47/57) of the electrodes had at least one tip within the cerebral cortex. Of the 47 electrodes, 28% (16/57) were in layer V and 55% (31/57) were in layer VI. In the remaining 17% (10/57), the tip was identified in either the external capsule or the corpus callosum (Figure 2C). The DV distribution of the electrode tips was comparable between sham and TBI rats (ለ<sup>2</sup> test; *p* > 0.05) (Supplementary Table S1).

*MRI cohort.* The DV location of the electrode tip was within the cortex in 76% (32/42) of rats, being in layer V in 31% (13/42) and layer VI in 45% (19/42) of the cases. In the remaining 24%, the tip was located in either the corpus callosum (7%, 3/42) or the external capsule (17%, 7/42) (Figure 2C). The DV distribution of the electrode tips was comparable between sham and TBI rats (ለ<sup>2</sup> test; *p* > 0.05).

#### *3.2. Success in Positioning the Posterior Intracortical Electrode*

#### 3.2.1. Electrode Locations—An Overview

*EEG cohort.* The electrode tip was within the cortex in 40% (21/53) of the rats. In the remaining 60% (31/53), it was either in the angular bundle, dorsal subiculum, or cortical lesion cavity (Figure 5 and Table 1). In four rats, the quality of the histological sections was not sufficient to determine the electrode location. The histological AP tip and ML electrode tip locations were within a 0.5-mm radius of the planned atlas coordinates in 19% (10/53) and 89% (47/53) of the rats, respectively (Figure 5A,B). The DV tip location was in the planned depth (layer V) in only 6% (3/53) of the rats. In the remaining 94% (50/53), it was either in layer VI, the angular bundle, dorsal subiculum, or cortical lesion cavity (Figure 5C).

**Figure 5.** Posterior intracortical electrode—schematic representations of the atlas-based, histological, and MRI-guided coordinates in each rat of the EEG and MRI cohorts. (**A**) Anteroposterior (AP) coordinate. In the EEG cohort (*n* = 47, electrode operation right after injury), the fixed atlas-based target AP coordinate of −7.56 mm from the bregma was applied to implant the electrodes (orange dots). In the MRI cohort (*n* = 40, electrode operation at 5 months postinjury), the target AP coordinate was individually determined using the in vivo 5-month T2-weighted MR images. The target coordinate fluctuated depending on the TBI (traumatic brain injury)-induced lesion extent. Note a mild anterior shift (*y*-axis) in the histologically verified "true" AP coordinate (blue dots) relative to the target

coordinate (orange dots) in both cohorts. In general, the anterior shift was less than that in a case of the anterior intra-cortical electrode (compare to Figure 1). Animal numbers are shown on the *x*-axis. (**B**) Mediolateral (ML) coordinate. In the EEG cohort (*n* = 47), the fixed atlas-based ML coordinate at 4 mm lateral to midline was targeted. In the MRI cohort (*n* = 40), the target ML coordinate was individually determined using the 5-month MRI. Note almost a negligible deviation of the histologically verified "true" ML coordinate from the atlas-based (EEG cohort) or MRI-guided (MRI cohort) coordinates. (**C**) Dorsoventral (DV) coordinate. In both cohorts, the lower tip of the bipolar electrode was targeted to layer V in the selected AP and ML coordinates (see above). In the EEG cohort, 46% (18/39), and in the MRI cohort, 66% (19/29) of the electrode tips in injured animals were in the cortex. Importantly, even though the lower tip in the remaining cases went down into the external capsule or corpus callosum, the upper tip of the bipolar electrode, being 1 mm higher in the EEG and 0.5 mm in the MRI cohort, was still recording in the cortex in 79% (31/39) of the rats in the EEG cohort and in 72% (21/29) in the MRI cohort. In 5 rats (3 sham, 2 TBI) in the EEG cohort and 10 rats (4 sham, 6 TBI) in the MRI cohort, the electrode was recording hippocampal rather than cortical activity, which affected the interpretation of the EEG data. The percentages of electrode locations in the sham-operated and TBI animals are shown on the right side of the panel. (**D**) Dot plots of the AP and ML shift in the histological AP and ML coordinate, and % of electrode in the targeted layer V (number of cases in brackets). Note posterior shift of some TBI cases from the target (Y = 0). The *y*-axis represents distance from target coordinate (Y = 0) or % of cases in targeted area. Note that in 4 animals in the EEG cohort and 2 in the MRI cohort, the DV location of the electrode tip could not be reliably determined in histological sections. Abbreviations: cavity, cortical lesion cavity; cc, corpus callosum; cg, cingulum; dcw, deep cerebral white matter; ec, external capsule; fmj, forceps major corpus callosum; HC, hippocampus; S, subiculum; V, ventricle.

*MRI cohort.* The electrode tip was within the cortex in 60% (24/40) of the rats. In the remaining 40% (16/40), it was either in the external capsule, corpus callosum, dorsal subiculum, hippocampus, or ventricle. In two rats, the quality of the histological sections was not sufficient to determine the electrode location. The histological AP and ML electrode tip locations were within a 0.5 mm radius of the planned MRI coordinates in 8% (3/40) and 100% (40/40) of the rats, respectively (Figure 5A,B and Table 1). The DV tip location was in the planned depth (layer V) in only 20% (8/40) rats. In the remaining 80% (32/40), the tip was located in either layer IV or VI, the external capsule, corpus callosum, dorsal subiculum, hippocampus, or ventricle (Figure 5C).

#### 3.2.2. Anteroposterior Location of the Posterior Intracortical Electrode Tips

Based on estimation of the progression of the cortical lesion, which tends to be more extensive caudally than rostrally [7,27,32], the posterior intracortical electrode was targeted to the atlas-based AP level −7.56 from the bregma (primary visual cortex, V1) to ensure proper EEG recording in the lesion vicinity during the 7th post-TBI month (Figure 1A).

*EEG cohort.* The unfolded cortical map indicated that all electrode tracks passed through the visual cortex. The electrode tips were in the visual cortex in only 40% (21/53) of the rats. In the remaining 60% (31/53), they were in the angular bundle, dorsal subiculum, or cortical lesion cavity (Figure 4A,B and Figure 5C). Typically, the tip location was anterior to the targeted atlas-based AP coordinate, the average deviation being 1.16 ± 0.09 mm (range: 0–2.28 mm, median: 0.96 mm) (Figure 5A,D). The rostral shift was comparable between the sham-operated controls and TBI rats (1.11 ± 0.17 mm vs. 1.18 ± 0.10 mm; *p* > 0.05) (V). The rostral shift of the posterior intracortical electrode was comparable to that of the anterior intracortical electrode (*p* > 0.05).

*MRI cohort.* As in the EEG cohort, histological analysis revealed that most of the posterior intracortical electrodes (60%, 24/40) were in the visual cortex. In the remaining 40% (16/40) of rats, the electrode track was in the visual cortex, but the tip penetrated either the deep cerebral white matter, cingulum, dorsal hippocampal commissure, hippocampus proper, or ventricle (Figure 4C,D and Figure 5C). In 95% (38/40) of the rats, the electrode tip was anterior to the targeted MRI-guided AP coordinate (Figure 5A,D

and Figure 6A,B,D,E), the average deviation being 1.39 ± 0.09 mm (range: 1.28–2.64 mm, median: 1.42 mm). The rostral shift was comparable between the sham-operated controls and TBI rats (sham 1.26 ± 0.18 vs. TBI 1.45 ± 0.12; *p* > 0.05) (Figure 5D). The rostral shift of the posterior intracortical electrode was comparable to that of the anterior intracortical electrode (*p* > 0.05).

**Figure 6.** Histological confirmation of the success of MRI-guided electrode placement. Left panel (**A**–**C**): Anterior intracortical electrode MRI-planned coordinate, histological confirmation, and "virtual" electrode. Right panel: Posterior intracortical electrode MRI-panned coordinate histological confirmation, and "virtual" electrode. (**A**) T2-weighted MRI and (**B**) histological images showing the MRI-guided (insert) and histology-confirmed "true" location of the anterior intracortical electrode in rat 1139. The anteroposterior (AP) coordinate was estimated by aligning the magnetic resonance images with the rat brain atlas [25]. The mediolateral (ML) and dorsoventral (DV) coordinates (inserts in (**A**,**D**)) were determined using ImageJ software (version 1.47v, Wayne Rasband and contributors, National Institute of Health, USA). Note that in this case, the confirmed AP location was about 1.8 mm more rostral than the planned location (−1.20 mm vs. −2.96). Lesion area is denoted in black-dashed-line circle. (**C**) A "virtual" location of the electrode tip (black line) if the electrode had been implanted to the targeted atlas-based coordinate (−1.75 mm from bregma, 4 mm from midline, 1.8 mm from the surface of the brain). (**D**) MRI-guided (insert) AP, ML and DV coordinates and (**E**) histology-confirmed "true" location of the posterior intracortical electrode tip in rat 1139. The black-dashed-line circle denotes the lesion area. Note that the confirmed AP location was approximately 1.4 mm more rostral than the planned location (−3.96 vs. −5.36). Thus, even though both the anterior and posterior intracortical electrodes were more rostral than planned, their tips were recording EEG signals in the perilesional cortex. (**F**) A "virtual" electrode (black line) at the atlas-based coordinates would have ended up in the lesion cavity. Scale bar in (**A**,**D**) = 1 mm, and in (**B**,**C**,**E**,**F**) = 2 mm.

#### 3.2.3. Mediolateral Location of the Posterior Intracortical Electrode Tips

*EEG cohort.* The histological ML tip coordinate was lateral to the targeted atlas-based coordinate in 62% (33/53), on target in 2% (1/83), and medial in 36% (19/53) of the rats (Figure 4B). The mean deviation from the atlas-based ML coordinate (0.38 ± 0.05 mm, range: 0–1.26 mm, median: 0.3 mm) was less than that of the anterior intracortical electrode (0.58 ± 0.05 mm, range: 0–1.50 mm; *p* < 0.05) (Figure 5D). The average deviation was comparable between sham-operated controls (0.45 ± 0.09 mm) and TBI rats (0.35 ± 0.05 mm; *p* > 0.05) (Figure 5D).

*MRI cohort.* The histological ML tip coordinate was lateral to the MRI-guided coordinate in 25% (10/40) and medial in 75% (30/40) of the rats (Figure 4B). The mean distance between the histological ML and MRI-guided coordinate was 0.45 ± 0.04 mm (range: 0.03–1.29 mm, median: 0.4 mm). The deviation was comparable between sham-operated controls (0.62 ± 0.09 mm) and TBI rats (0.39 ± 0.04 mm; *p* > 0.05) (Figure 5D). Also, the deviation from the MRI-based ML did not differ from that of the anterior intracortical electrode (0.44 ± 0.04 mm; *p* > 0.05) (Figure 5D).

#### 3.2.4. Dorsoventral Location of the Posterior Intracortical Electrode Tips

*EEG cohort.* The histologically verified DV location of the lower tip of the posterior intracortical electrode was more diverse compared with the anterior electrode, with only 40% (21/52) within the cortical layers (6% in layer V, 34% in layer VI) (Figure 5C and Table 1). Moreover, 40% (21/52) of the tips had traveled down to the angular bundle, 10% (5/53) to the dorsal subiculum, and 10% (5/52) to the underlying cortical lesion cavity (Figure 5C).

Further analysis indicated that the laminar location of the electrode tips differed between the posterior and anterior electrodes (ለ<sup>2</sup> test; *p* < 0.001). The overall percentage of posterior electrode tips in the cortex was less than that of the anterior electrode (40% vs. 83%; *p* < 0.001) (Table 1). Also, fewer posterior electrode tips were located in the targeted layer V compared with anterior cortical electrode tips (9% vs. 28%; *p* < 0.01) (Figure 5C, Supplementary Table S1). Surprisingly, the overall percentage of posterior cortical electrodes in the cortex was comparable between sham (46%; 5/11) and TBI rats (67%, 19/29; *p* > 0.05). Also, there was no difference in the overall DV distribution of the locations of the posterior intracortical electrode tips between sham and TBI rats (ለ<sup>2</sup> test; *p* > 0.05) (Supplementary Table S1).

*MRI cohort.* Altogether, 60% (24/40) of the posterior intracortical electrodes had the lower tip within the cortex, of which 2% (1/40) were in layer IV, 20% (8/40) in layer V, and 38% (15/40) in layer VI (Figure 5C and Table 1). The tip had reached the external capsule or cingulum in 8% (4/40), the angular bundle or hippocampus proper in 27% (11/40), and the ventricle in 5% (2/40). Unlike in the EEG cohort, none of the electrode tips appeared to enter the lesion cavity (Figure 5C). The overall DV tip distributions did not differ between sham-operated controls and TBI rats (ለ<sup>2</sup> test; *p* > 0.05) (Supplementary Table S1).

The overall percentage of tips in the cortex was comparable between the anterior and posterior cortical electrodes (76% vs. 60%, ለ<sup>2</sup> test; *p* > 0.05), and between sham and TBI rats (46% vs. 66%, ለ<sup>2</sup> test; *p* > 0.05) (Table 1). Like in the EEG cohort, however, the overall DV distribution of the electrode tips differed between the anterior and posterior intracortical electrodes (ለ<sup>2</sup> test; *p* < 0.01). Particularly, the percentage of electrodes entering into the septal hippocampus was greater in the TBI rats than in the sham group (27% vs. 0%, ለ<sup>2</sup> test; *p* < 0.001).

#### *3.3. Success in Positioning the Intrahippocampal Electrode*

#### 3.3.1. Electrode Locations—An Overview

The lower tip of the hippocampal electrode was targeted to the hilus of the septal end of the dentate gyrus (AP: −3.0; ML: −1.4; DV: 3.6) in both the EEG and MRI cohorts.

*EEG cohort.* The electrode tip was in the hippocampus or dentate gyrus in 82% (47/57) of the rats (Table 1). In the remaining 18% (10/57), the tip was located in either the

fimbria, ventricle, dorsal thalamus, or out of the hippocampus (unidentified tip location) (Figure 7). The histologically verified AP and ML electrode tip locations were within 0.5 mm of the planned atlas coordinates in 65% (37/57) and 100% (56/56) of rats, respectively (Figure 7A,B). In one rat, the quality of the histological sections was not sufficient to determine the electrode location. The DV tip location was in the dentate gyrus in 54% (31/57) of rats. In the remaining 46% (26/57), the tip was located in either the CA1 or CA3 subfield of the hippocampus proper, fimbria, ventricle, or dorsal thalamus (Figure 7C).

**Figure 7.** Hippocampal electrode—schematic representations of the atlas-based, histological, and

MRI-guided coordinates in each rat of the EEG and MRI cohorts. (**A**) Anteroposterior (AP) coordinate. In the EEG cohort (electrode operation right after injury), the fixed atlas-based target AP coordinate of −3 mm from the bregma was applied to implant the electrodes (orange dots). In the MRI cohort (electrode operation 5 months after injury), the target AP coordinate was individually determined using the 5-month in vivo T2-weighted MR images. The target coordinate fluctuated, depending on the TBI-induced hippocampal structural abnormality. Note the anterior shift (*y*-axis) in the histologically verified "true" AP coordinate (blue dots) relative to the aimed target coordinate (orange dots) in both cohorts. Note the great variability in the anterior shift from animal to animal, particularly in the MRI cohort. Animal numbers are shown on the *x*-axis. (**B**) Mediolateral (ML) coordinate. In the EEG cohort, the fixed atlas-based target ML coordinate at 1.4 mm lateral to midline was targeted. In the MRI cohort, the target ML coordinate was individually determined using the 5-month MRI. Note only a very small deviation of the histologically defined "true" ML coordinate from the atlas-based (EEG cohort) or MRI-guided (MRI cohort) coordinates. (**C**) Dorsoventral (DV) coordinate. In both cohorts, the lower tip of the bipolar electrode was aimed at the hilus in the selected AP and ML coordinates (see above). In the EEG cohort, most of the tips were recording in the hippocampus proper or the dentate gyrus. In the EEG cohort, in only 19% (8/43) of TBI cases, the tip was either in fimbria, ventricle, or went through the septal hippocampus to the dorsal thalamus or to an unidentified location. In the MRI cohort, in only 14% (6/31) of TBI cases, the tip was outside the hippocampus or the dentate gyrus. The percentages of electrode locations in the sham-operated and TBI animals are shown on the right side of the panel. (**D**) Dot plots of the AP and ML shift in the histological AP and ML coordinate, and % of electrodes in the targeted dentate gyrus (number of cases in brackets). Note posterior shift of some cases from the target (vertical dashed line). The *y*-axis represents distance from target coordinate (Y = 0) or % of cases in targeted area. Abbreviations: alv, alveus; CA1, CA1 subfield of the hippocampus; CA3, CA3 (CA3b, CA3c) subfield of the hippocampus; gcl, granule cell layer (s-gcl, suprapyramidal blade, i-gcl, infrapyramidal blade); hf, hippocampal fissure; l-m, stratum lacunosum moleculare of CA1; mol, molecular layer of the dentate gyrus; V, ventricle.

*MRI cohort.* The electrode tip was in the hippocampus proper or dentate gyrus in 86% (36/42) of rats (Table 1). In the remaining 14% (6/42), the tip was located in either the ventricle or fimbria (Figure 7). The histologically verified AP and ML electrode tip locations were within 0.5 mm of the planned atlas coordinates of the MRI-guided location in 14% (6/43) and 100% (42/42) of rats, respectively (Figure 7A,B). The electrode tip was in the planned DV coordinate in 38% (16/42) of the rats. In the remaining 62% (26/42), the tip was located in either the hippocampal CA1 or CA3 subfields, hippocampal fissure, fimbria, or ventricle (Figure 7C).

#### 3.3.2. Anteroposterior Location of Hippocampal Electrode Tips

*EEG cohort.* Only 9% (5/57) of the electrode tips were in the planned atlas-based coordinate (AP: 3.0 mm). The remaining 77% (44/57) were positioned rostrally and 14% (8/57) were positioned caudally (Figure 7A). The mean deviation of the histologically verified coordinate from the atlas-based coordinate was 0.57 ± 0.07 mm (range: 0–2.52 mm, median: 0.48 mm). The deviation was comparable between sham-operated controls and TBI rats (0.54 ± 0.09 mm vs. 0.58 ± 0.09 mm; *p* > 0.05) (Figure 7D).

*MRI cohort.* The AP distribution of MRI-guided hippocampal electrode tip placements was anterior in 93% (39/42) and posterior in 7% (2/42) of the cases (Figure 7A). The average deviation of the histologically verified coordinate from the MRI-guided coordinate was 0.95 ± 0.08 mm (range: 0.04–2.28, median: 0.84) (Figure 7D). The deviation was comparable between sham-operated controls and TBI rats (0.66 ± 0.11 mm vs. 1.06 ± 0.10 mm; *p* > 0.05) (Figure 7D).

#### 3.3.3. Mediolateral Location of Hippocampal Electrode Tips

*EEG cohort.* The hippocampal electrode was located anterior to the atlas-defined target in 48% (27/56), medial in 50% (28/56), and on target in 2% (1/56) of the rats (Figure 5B). The mean deviation of the histologically defined coordinate from the atlas-defined coordinate was 0.21 ± 0.02 mm (range: 0–0.63 mm, median: 0.2 mm) (Figure 7D). The deviation was comparable between sham-operated controls and TBI rats (0.24 ± 0.04 mm vs. 0.19 ± 0.03 mm; *p* > 0.05) (Figure 7D).

*MRI cohort.* The histologically defined tip coordinate was lateral to the MRI-guided coordinate in 17% (7/42) and medial in 83% (34/42) of the rats. The average deviation of the histologically defined coordinate from the MRI-guided coordinate was 0.34 ± 0.04 mm (range: 0–0.9 mm, median: 0.3 mm) (Figure 7D). The deviation was comparable between sham-operated controls and TBI rats (0.35 ± 0.07 mm vs. 0.34 ± 0.05 mm; *p* > 0.05) (Figure 7D).

#### 3.3.4. Dorsoventral Location of Hippocampal Electrode Tips

*EEG cohort.* Histology revealed electrode path-associated lesions (see details in Section 3.4.2). In some TBI rats, the electrode path appeared slanted rather than vertical, probably due to hippocampal distortion related to enlarged ventricles.

The electrode tip was in the dentate gyrus in 49% (28/57) of the rats, locating in the molecular layer of the dentate gyrus in 17% (10/57) and in the granule cell layer in 32% (18/57) of the cases. No tips were observed in the hilus (Figure 7C). In the remaining rats, the tip was located in the hippocampus proper in 33% (5% [3/57] in the stratum lacunosum moleculare or hippocampal fissure, 5% [3/57] in CA1, 23% [13/57] in CA3), most of the tips in the CA3 subfield being in the CA3c subfield (Figure 5C). In 9% (5/57) of the cases, the tip was in the fimbria or ventricle. In 9% (5/57), the electrode went through the septal hippocampus and ended in the dorsal thalamus (*n* = 2) or in an unidentified location (*n* = 3) (see details in Section 3.4.2). The DV distribution of the electrode tips was comparable between sham-operated controls and TBI rats (ለ<sup>2</sup> test; *p* > 0.05).

*MRI cohort.* Unlike in the EEG cohort, electrode path-associated lesions were small. Also, the electrode paths were vertical rather than slanted.

The electrode tip was in the targeted dentate gyrus in 36% (15/42) of animals, being 8% (3/42) in the molecular layer and 28% (12/42) in the granule cell layer (Figure 7C). No electrode tips were observed in the hilus. In 50% of the rats, the tip was in the hippocampus proper (7% [3/42] in the stratum lacunosum moleculare or hippocampal fissure, 5% [2/42] in CA1, and 38% [16/42] in CA3). Like in the EEG cohort, most of the tips in the CA3 were in the CA3c subfield (Figure 7C). In the remaining 14% (6/42), the tip was located in either the fimbria (*n* = 1) or the ventricle (*n* = 5) (Figure 7C). The DV distribution of tip locations was comparable between sham-operated controls and TBI rats (ለ<sup>2</sup> test; *p* = 0.741).

#### *3.4. Cortical and Hippocampal Atrophy after TBI*

#### 3.4.1. Cortical Atrophy

*Anteroposterior.* In sham-operated controls, there was no difference in the cortical AP length at 4 mm lateral to the midline between the ipsilateral (13.89 ± 0.19 mm, range: 12.48–0.56 mm, median: 14.1 mm) and contralateral hemispheres (13.70 ± 0.09 mm, range: 13.12–14.24 mm, median: 13.6 mm; *p* > 0.05) (Figure 8A1,B,C).

In TBI rats, the cortical AP length was shorter ipsilaterally (12.37 ± 0.16 mm, range: 10.08–13.76 mm, median: 12.23 mm) than contralaterally (13.13 ± 0.17, range: 11.04–14.72 mm, median: 13.12 mm; *p* < 0.001) (Figure 8A2–A4,B). Both the ipsilateral (*p* < 0.001) and contralateral (*p* > 0.05) AP lengths were shorter in the TBI rats than in the sham group (Figure 8B,C).

*Mediolateral.* In sham-operated controls, the ML length was similar ipsilaterally (5.62 ± 0.06 mm, range: 5.28–6.08 mm, median: 5.6 mm) and contralaterally (5.51 ± 0.06 mm, range: 5.12–5.76, median: 5.6 mm; *p* > 0.05).

In the TBI group, the ML length was shorter ipsilaterally (5.03 ± 0.03 mm, range: 4.64–5.44, median: 4.96 mm) than contralaterally (5.27 ± 0.04 mm, range: 4.80–5.92 mm, median: 5.28 mm; *p* < 0.001). Both the ipsilateral (*p* < 0.001) and contralateral (*p* < 0.05) ML lengths were shorter in the TBI rats than in the sham group (Figure 8D).

**Figure 8.** Anteroposterior and mediolateral shrinkage of the brain. In vivo magnetic resonance 3D multigradient echo (MGRE) images acquired at 5 months after TBI were used to estimate cortical shrinkage in the MRI cohort. (**A1**–**A4**) coronal, sagittal (ipsilateral and contralateral) and horizontal MGRE images of a sham rat (**A1**) and TBI rats (**A2**–**A4**). Anteroposterior (AP) cortical shrinkage was estimated by measuring the distance between the rostral and caudal cortical surface (double-headed arrows) in the sagittal slice at 4 mm from the midline both ipsilaterally (orange) and contralaterally (white). Note the change in the shape of the ipsilateral cortex (sagittal images) in TBI rats, indicating the TBI-induced cortical atrophy (see also turquoise arrows in (**A3**,**A4**)). Mediolateral (ML) shrinkage was assessed by measuring the distance between the midline and the lateral edge of the cortex (turquoise double headed arrow) in a horizontal slice at 1.7 mm below the pial surface at AP level −1.56 (corresponding to the targeted location of the anterior intracortical electrode tip). (**B**) A dot plot showing the ipsilateral (orange) and contralateral (blue) cortical AP lengths (*y*-axis) in the sham and TBI groups (*x*-axis). Note that both the ipsilateral and contralateral cortical AP lengths were reduced

in TBI rats compared with sham-operated animals. Also, in the TBI group, the cortical AP length was shorter ipsilaterally than contralaterally. (**C**) A paired dot plot showing that the ipsilateral vs. contralateral shrinkage in each rat. The greater the ipsilateral shrinkage, the greater the contralateral shrinkage in the TBI compared with sham group. Arrows point to the 3 cases illustrated in panels (**A1**– **A4**). (**D**) A dot plot showing the ipsilateral and contralateral cortical ML lengths in sham-operated and TBI rats. Note that both the ipsilateral and contralateral cortical ML lengths were reduced in TBI rats compared with sham-operated animals. Also, in the TBI group, the cortical ML length was shorter ipsilaterally than contralaterally. Statistical significance: \*\*\* *p* < 0.001 compared with the contralateral hemisphere (Wilcoxon signed-rank test); ### *p* < 0.001, # *p* < 0.05 compared with the sham group (Mann–Whitney *U* test).

#### 3.4.2. Hippocampal Atrophy

*Anteroposterior.* In the sham group, the average AP length was similar ipsilaterally (7.40 ± 0.12 mm, range: 6.88–8.0 mm, median: 7.36 mm) and contralaterally (7.36 ± 0.09 mm, range: 7.04–8.0 mm, median: 7.36 mm; *p* > 0.05) (Figure 9D1,E,F).

**Figure 9.** (**A**–**C**) Electrode tracts. Histological images from the coronal thionine-stained sections of 3 rats,

showing the tracts of the bipolar intracortical electrodes and locations of the lower electrode tip (filled arrowhead). The target of the lower tip was the hilus. In panel (**A**), the electrode tip is in the suprapyramidal blade of the granule cell layer. In panel (**B**), the tip went through the dentate gyrus down to the dorsal thalamus. In panel (**C**), the tip is in the infrapyramidal blade of the granule cell layer. Note the electrode-path associated lesion in CA1 (open arrow). (**D1**–**D3**) Coronal, sagittal and horizontal in vivo magnetic resonance 3D multigradient echo (MGRE) images of the ipsilateral and contralateral hippocampus were used to assess hippocampal shrinkage after traumatic brain injury (TBI). (**D1**–**D3**) Hippocampal distortion and shrinkage. Panel (**D1**): A sham-operated experimental control (1107). Panels (**D2**,**D3**): Two rats with TBI (1028, 1144). The anteroposterior (AP) shift of the hippocampus was assessed by measuring the distance from the rostral edge of the frontal cortex to the rostral edge of the hippocampus at 1.4 mm from the midline in the horizontal slice 2.8 mm below the surface of the brain (left hemisphere: orange double-headed arrow; right hemisphere: white double-headed arrow). Mediolateral (ML) shrinkage was assessed by measuring the distance from the brain midline to the lateral edge of the hippocampus in the same horizontal plane (2.8 mm below the surface of the brain, turquoise double-headed arrow) in a slice sampled at AP level −2.8 mm, corresponding to the AP level of the atlas-based target coordinate. In both TBI rats (**D2**,**D3**), the distance from the frontal pole to the rostral edge of the hippocampus was longer than that in the sham-operated animal, indicating retraction of the septal hippocampus caudally. The ML length in TBI rats (**D2**,**D3**) was shorter than that in the sham-operated animal (**D1**), indicating a shift toward midline. (**E**) A dot plot showing the ipsilateral (orange) and contralateral (blue) anteroposterior lengths (*y*-axis) in the sham and TBI groups (*x*-axis). Note that both the ipsilateral and contralateral cortical AP lengths were increased in TBI rats compared with sham-operated animals. Also, in the TBI group, the cortical AP length was greater ipsilaterally than contralaterally. (**F**) A paired dot plot showing the ipsilateral vs. contralateral backward "movement" in each rat. The greater the ipsilateral "movement", the greater the contralateral "movement". Arrows point to the 3 cases illustrated in panels (**D1**–**D3**). (**G**) A dot plot showing the ipsilateral and contralateral hippocampal ML lengths in sham-operated and TBI rats. Note that both the ipsilateral and contralateral cortical ML lengths were reduced in TBI rats compared with sham-operated animals. Also, in the TBI group, the ML length was shorter ipsilaterally than contralaterally. Statistical significance: ##, *p* < 0.05; ###, *p* < 0.001 as compared to the sham group (Mann–Whitney U test); \*\*\* *p* < 0.001 compared with the contralateral hemisphere (Wilcoxon signed-rank test). Scale bar in (**A**–**C**) = 500 μm.

In the TBI group, the AP length was greater ipsilaterally (8.16 ± 0.09 mm, range: 7.2–8.96 mm, median: 8.16 mm) than contralaterally (7.83 ± 0.07 mm, range: 6.88–8.64 mm, median: 7.84 mm; *p* < 0.001) (Figure 9D2,D3,E,F). Both the ipsilateral (*p* < 0.001) and contralateral (*p* < 0.01) AP lengths were greater in the TBI rats than in the sham group.

*Mediolateral.* In the sham group, the ML length was comparable ipsilaterally (4.09 ± 0.05 mm, range: 3.68–4.32 mm, median: 4.16 mm) and contralaterally (3.83 ± 0.06 mm, range: 3.52–4.0 mm, median: 3.84 mm; *p* > 0.05) (Figure 9D1,G).

In the TBI group, the ML length was shorter ipsilaterally (2.48 ± 0.08 mm, range: 1.92–3.84 mm, median: 2.4 mm) than contralaterally (3.53 ± 0.04 mm, range: 3.04–4.0 mm, median: 3.52 mm; *p* < 0.001) (Figures 7G and 9D2,D3). Both ipsilateral (*p* < 0.001) and contralateral (*p* < 0.01) ML lengths were shorter in the TBI rats than in the sham group (Figure 9).

#### *3.5. "Virtual Electrode"—Comparison of Success Rate in Atlas-Based vs. MRI-Guided Electrode Placements*

Finally, to assess whether the MRI images indeed improved the targeting of the electrode tip to the perilesional cortex, and not, for example, to the lesion cavity, we reexamined the histological sections of TBI rats in the MRI cohort. We focused on the caudal aspect of the brain, as targeting this area without the use of MRI was challenging due to remarkable TBI-related cortical atrophy.

A hypothetical "virtual" electrode was placed at the atlas-defined coordinate of the posterior intracortical electrode (Figure 10). We then reconstructed the destination of the electrode tip in the available histological sections by assessing (a) whether it was located in the cortex or lesion cavity and (b) the distance of the tip from the lesion edge. We found that by using the atlas-based coordinate (AP −7.56), 58% (18/31) of the electrodes had been in the lesion cavity compared with 0% for the MRI-guided implantations (Figures 5C and 10 and Table 2). The remaining 42% (13/31) of the "virtual" electrodes were located medial to the lesion cavity, except in one case (rat 1103), in which the tip location was caudal to the lesion. The average distance of the electrode tip to the lesion edge was 0.64 ± 0.1 mm (range: 0–1.3 mm) (see also Supplementary Table S2 for further details).

**Figure 10.** Location of electrode tip at 5 months postinjury without prior MRI analysis. Photomicrographs of thionine-stained coronal brain sections of 4 animals; (**A**) #1019, (**B**) #1139, (**C**) #1158 and (**D**) #1036 in the MRI cohort with electrode implantations at 5 months after TBI. Left panels: MRI-guided placement of the posterior cortical electrode. Note that all electrodes are within the perilesional cortex. Right panels: The location of the electrode tip (arrow), if the electrode was implanted according to the targeted atlas-based coordinates (−7.56 mm from bregma, 4 mm from midline, 1.8 mm from the surface of the brain). Note that in all rats except 1019, the electrode tip ended in the lesion cavity. Table 2 summarizes the locations for all cases. Scale bar = 2 mm.

**Table 2.** Location of the "virtual electrode". Summary of the locations of the posterior intracortical electrodes in the MRI cohort, if implanted according to the atlas-based coordinates. Note that 58% (18/31) of the lower electrode tips were in the cortex while 42% (13/31) of the lower tips were in the lesion cavity. After MRI-guidance, 71% (22/31) of the lower tips were recording in the cortex. In 5 additional cases, the upper tip (0.5 dorsal to the lower tip) was expected to record in the cortex, resulting in a total of 87% (27/31) of the electrodes recording in the cortex. Only 1 electrode was not recording in the brain.


#### **4. Effect of Electrode Implantation on Progression of the Cortical Lesion**

Next, we assessed whether a 6-month-long presence of intracortical electrodes enhanced cortical atrophy. We hypothesized that (1) the lesion area would be greater in the EEG cohort than in the MRI cohort and (2) the cortical electrode tips would be closer to the lesion cavity (expected distance ≥ 500 μm) in the EEG cohort than in the MRI cohort.

*Lesion area and location.* The cortical lesion spread laterally and caudally, typically involving the sensory, auditory, and visual cortices, as previously described [32] (Figure 11A). The average total lesion area was comparable between the EEG (27.31 ± 2.29 mm2, range: 1.72–93.8 mm2, median: 25.8 mm2) and MRI (25.69 ± 2.32 mm2, range: 5.7–47.6 mm2, median: 26.1 mm2) cohorts (*p* > 0.05) (Figure 11B). Also, the mean lesion area in the primary somatosensory (4.89 ± 0.41 mm<sup>2</sup> vs. 5.16 ± 0.55 mm2; *<sup>p</sup>* > 0.05) and visual cortex (7.73 ± 0.72 mm2 vs. 7.37 ± 0.51 mm2; *<sup>p</sup>* > 0.05) was similar in the EEG and MRI cohorts (Table 3). In the secondary somatosensory cortex (S2) (0.56 ± 0.07 mm2 vs. 1.07 ±0.17 mm2; *<sup>p</sup>* < 0.01)

and the primary auditory cortex (Au1) (3.77 ± 0.20 mm<sup>2</sup> vs. 4.32 ± 0.30 mm2; *<sup>p</sup>* < 0.05) the lesion area was smaller in the EEG cohort than in the MRI cohort (Table 3).

**Figure 11.** Distance of the intracortical electrodes from the edge of the cortical lesion cavity. (**A**) An

unfolded cortical map of a rat 1064, showing the cytoarchitectonic distribution of the cortical lesion (blue outline) and the location of the anterior (brown filled circle in the S1BF) and posterior (yellow filled circle in the V2L) intracortical electrodes. Note that the lesion had progressed laterally and caudally. Consequently, the posterior electrode was closer to the lesion cavity edge than the anterior electrode. (**B**) A scatter plot showing the cortical lesion area in the EEG and MRI cohorts (each dot represents 1 rat). The lesion area was comparable between cohorts (*p* > 0.05). (**C**) In the EEG cohort, the distance from the electrode tip to the lesion cavity edge (layer V intersection was used as reference) was similar between the anterior and posterior intracortical electrodes (*p* > 0.05). (**D**) In the MRI cohort, the distance from the anterior intracortical electrode tip to the cavity edge was slightly greater than that from the EEG cohort (*p* < 0.05). (**E**) In the EEG cohort, the larger the lesion, the closer the posterior electrode tip to the lesion cavity edge (*p* < 0.001). (**F**) In the MRI cohort, the larger the lesion, the closer the posterior electrode tip to the cavity edge (*p* < 0.001). Abbreviations: S1BF, primary somatosensory barrel field; V2L, secondary visual cortex lateral area. Statistical significance: #, *p* < 0.05 compared with the EEG cohort (Mann–Whitney U test).

*Electrode distance from the lesion cavity.* In the EEG cohort, 54% of the anterior intracortical electrodes were located anterior to the rostral edge of the lesion and 46% were located medial to the lesion. The average distance to the lesion edge was 0.79 ± 0.08 mm (range: 0–2.36 mm, median: 0.71 mm) (Figure 11C). All posterior intracortical electrodes were located medial to the lesion; 33%, however, were observed on the edge of the lesion or in the lesion cavity. The mean distance to the lesion edge was 0.63 ± 0.11 mm (range: 0–2.17 mm, median: 0.38 mm), comparable to that of the anterior intracortical electrode (*p* > 0.05).

In the MRI cohort, 41% of the anterior intracortical electrodes were located anterior to the rostral edge of the lesion and 59% were located medial to the lesion. The distance to the lesion edge was greater in the MRI (1.038 ± 0.10 mm, range: 0–2.3 mm, median: 1.03 mm) than in the EEG cohort (*p* < 0.05). All posterior intracortical electrodes (100%) were located medial to the lesion. Unlike in the EEG cohort, none of the tips was at the lesion edge or in the cavity. The distance of the posterior electrode tip to the lesion edge (0.79 ± 0.15 mm, range: 0–2.7 mm, median: 0.54 mm) was comparable to that of the anterior intracortical electrode (*p* > 0.05) (Figure 11D) or the posterior intracortical electrode in the EEG cohort (*p* > 0.05).

In both cohorts, we were unable to verify the hypothesis that the greater the lesion, the closer the tip of the anterior intracortical electrode to the lesion edge as there was no correlation between lesion size and the distance of the anterior intracortical electrode from the lesion edge (EEG cohort: R = −0.237; *p* > 0.05; MRI cohort: R = −0.288; *p* > 0.05) (Figure 9E,F). In the case of posterior cortical electrodes, however, the larger the lesion, the closer the tip to the lesion edge (EEG cohort: R = −0.712; *p* < 0.0001, MRI cohort: R = −0.411; *p* < 0.05) (Figure 9E,F).


**Table 3.** Cytoarchitectonic distribution of the TBI-induced cortical lesion in the EEG and MRI cohorts.

Abbreviations: Au1, primary auditory cortex; AuD, secondary auditory cortex, dorsal area; AuV, secondary auditory cortex, ventral area; DLEnt, dorsolateral entorhinal cortex; DI, dysgranular insular cortex; Ect, ectorhinal cortex; GI, granular insular cortex; LPtA, lateral parietal association cortex; MPtA, medial parietal association cortex; PRh, perirhinal cortex; PtPD, parietal cortex, posterior area, dorsal part; PtPR, parietal cortex; RSD, retrosplenial dysgranular cortex; RSGc, retrosplenial granular cortex, c region; S1, primary somatosensory cortex; S2, secondary somatosensory cortex; S1BF, primary somatosensory cortex, barrel field; S1DZ, primary somatosensory cortex, dysgranular zone; S1FL, primary somatosensory cortex, forelimb region; S1Sh, primary somatosensory cortex, shoulder region; S1Tr, primary somatosensory cortex, trunk region; S1ULp, primary somatosensory cortex, upper lip region; TeA, temporal association cortex; V1, primary visual cortex; V1M, primary visual cortex, monocular area; V1B, primary visual cortex, binocular area; V2L, secondary visual cortex, lateral area; V2ML secondary visual cortex, mediolateral area; V2MM, secondary visual cortex, mediomedial area. Area is shown as a mean ± standard error of the mean (SEM). Animal numbers are in parenthesis. Statistical significances: \*, *p* < 0.05 as compared to the area in the EEG cohort (Mann-Whitney U); #, *p* < 0.05 as compared to the percentage of rats in the EEG cohort (ለ<sup>2</sup> test).

#### **5. Discussion**

To address the challenges related to chronic implantation of electrodes in braindamaged rats, our objective was to develop methodologies to maximize the accuracy of chronic recording-electrode placements using preimplantation structural MRI. In the material available, we also assessed the effect of chronic electrode implantations on TBI- induced brain atrophy. Our data revealed that (1) animal-dependent progression of the cortical lesion after TBI compromises the placement accuracy of depth electrodes implanted at later time-points when only atlas-based coordinates are used; (2) MRI-guided adjustment of atlas-based coordinates increases the placement accuracy of intracerebral electrodes at the chronic post-TBI phase, particularly in the perilesional cortex; and (3) chronically implanted electrodes do not increase cortical and/or hippocampal atrophy.

#### *5.1. Electrode Implantation Immediately after TBI Resulted in Good Location Accuracy of Intracerebral Electrode Tips Rostrally, but Was Less Accurate Caudally*

In the EEG cohort, we implanted two bipolar intracortical and one bipolar hippocampal electrode ipsilateral to the lesion immediately after TBI to monitor the acute post-TBI electrophysiologic events and followed up the evolution of epileptiform activities over the following 7 months [36,37]. We assumed that the deformation and atrophy of the brain were not compromising the electrode placement accuracy at this early postinjury time-point, and thus we could rely on the rat brain atlas designed for the normal brain in defining the AP and DV coordinates for electrode placements. The fixed atlas-based cortical and hippocampal AP coordinates were chosen based on our previous observations that the cortical lesion progresses laterally and caudally [32]. In particular, it is critical to position the electrode tip in the anticipated epileptogenic area in the perilesional cortex, but avoid the lesion cavity [13].

Despite the unpredictable caudal progression of the cortical lesion, 83% of the anterior and 40% of the posterior cortical electrodes were within the cortex, and there was no difference in the overall DV distribution of the electrode tips between sham-operated controls and TBI rats. In 10% of the animals, however, the electrode tip of the posterior cortical electrodes had entered the lesion cavity when assessed at 7 months after implantation. This was particularly evident in cases with a large lesion size, which associated with robust cortical thinning. Also, 10% of the posterior cortical electrodes had entered the hippocampus, and consequently, recorded hippocampal EEG instead of cortical EEG. Progression of cortical lesions in the lateral FPI model is known to be variable and unpredictable [7,21,32,38]. It is possible that the progressive cortical atrophy "melting" of the brain around the electrode tips may have caused many of the electrode tips to end in the lesion cavity or subcortical areas, including the angular bundle and hippocampus. Even in the presence of an acute preimplantation MRI of the rats included in the EEG cohort, it was difficult to estimate the correct location for the intracerebral cortical electrodes 7 months later.

Although we were quite successful in implanting the electrode tips in the cortical tissue, both the anterior and posterior electrodes tended to locate slightly more anterior than planned. The anterior deviation from the target was typically less than 1 mm, however, and the mediolateral deviation was even smaller. We assume that the use of the Sprague Dawley strain in the present experiments instead of Wistar rats, which were used in the atlas preparation, and also mild human errors in reading bregma, could have affected the accuracy of the electrode placements. It is also important to note that depending on the skull size, the location of the 5 mm-diameter craniotomy slightly varied, which could also affect electrode insertion.

The target of the lower hippocampal electrode tip was at the hilus of the dentate gyrus. Consequently, the upper tip was expected to be in the CA1 subfield. As such, we aimed to have at least one electrode tip recording hippocampal epileptiform activity. Although in previous studies we noticed the development of hippocampal atrophy and deformations over months postinjury, we had no quantitative data to estimate the adjustments needed to fix the atlas-guided electrode tip coordinates. Somewhat disappointingly, at 7 months after the electrode implantation in the EEG cohort, the electrode tip was outside the dentate gyrus in half of the cases. In the remaining cases, the electrode tips were mainly in the CA3c subfield of the hippocampus proper. Only 7% of the electrodes were outside the hippocampus and not recording. In two cases, the electrodes were recording in the thalamus. We assume that postimpact subdural hematoma and/or edema affected the reading of the

pial surface, which is used to calculate the DV coordinate, leading to mild misplacement of the hippocampal electrode in the TBI rats, particularly as the CA3c misplacement was observed in 1 sham rat only. Moreover, the implantation was technically challenging, as the AP coordinate of the hippocampal electrode was very close to the rostromedial edge of the craniotomy, leading in some cases to more rostral electrode repositioning of the AP coordinate planned during the surgery.

Taken together, our data show that the location of intracortical electrodes implanted immediately after TBI in the lateral FPI model, particularly to the posterior parts of the brain, can be compromised by unpredictable caudolateral progression of the cortical lesion and cortical thinning, leading to electrode misplacement affecting the EEG recordings. The hippocampal electrode placements immediately after TBI for chronic EEG followup are generally more stable and less affected by post-TBI hippocampal morphologic transformations; even the acute cortical swelling and subdural hematoma can affect DV electrode placement. As 30% of the 297 intracortical or hippocampal electrodes were recording outside the target tissue (cortex or hippocampus/dentate gyrus), verification of the electrode tip locations is needed for accurate interpretation of chronic EEG recordings even when the electrode implantations are performed in the early postinjury time period.

#### *5.2. Preimplantation Structural MRI Improved the Accuracy of Electrode Implantations at 5 Months Postinjury*

The interim review of MRIs imaged during the 5-month follow-up of animals in the MRI cohort confirmed our previous observations of the progression of cortical lesion in injured rats and raised concerns about the accuracy of electrode placements for the 1-month 24 h/7-day high-density video-EEG recordings, which was critical for the epilepsy phenotyping in our animal cohorts. It is generally recognized that proper electrode placement is a basis for accurate EEG data interpretation. To maximize the success rate of electrode implantations, we used images of coronal in vivo MR T2-wt slices obtained at 5 months post-TBI to generate rat-specific AP, ML, and DV coordinates for the anterior and posterior cortical electrodes, as well as the hippocampal electrodes. The aim was to place the cortical electrodes in the perilesional cortex within 500 μm of the lesion cavity, similar to that in the EEG cohort. Moreover, we wanted to get the DV coordinate of the ventral electrode tip to layer V of the sometimes very atrophied cortex. As in the EEG cohort, the hippocampal electrode tip was targeted to the dentate gyrus.

Analysis of histological sections prepared from the same rats approximately 1 month after the MRI, i.e., right after finishing the 1-month video-EEG monitoring, revealed a rostral shift of both the cortical and hippocampal electrodes relative to the MRI-guided coordinate. These data suggest that cortical atrophy, retracting the brain backwards may have contributed to the anterior shift and low accuracy of the cortical electrodes, particularly the posterior cortical electrode, in injured animals [7,17,39]. In case of hippocampal electrodes, the anterior shift was clearer in TBI rats than in sham animals, similar to the EEG cohort. Even though deviation of the hippocampal electrodes from the planned position can also relate to hippocampal atrophy, changes in its septotemporal orientation, rotation, and medial shift toward midline can contribute to electrode misplacements [23].

Generally, despite the unexpected divergence between the MRI-defined and histologically verified "true" coordinates, our data demonstrate that the MR images were useful when targeting the perilesional cortex for EEG recordings during the chronic phase post-TBI. Our "simulation" revealed that most (60%) of the MRI-guided posterior cortical electrode placements were in the cortex compared with 42% of the posterior cortical electrode placements when only the atlas-based coordinate was used. Moreover, with MRI-guided implantation, we were able to avoid the cortical lesion cavity. Additionally, the effect of cortical thinning on the DV location of the posterior electrode was mitigated, as the percentage of electrodes in the cortex was comparable between sham and TBI rats. Moreover, there was a 40% increase in cases with a cortical electrode location and a 14% increase in reaching the target layer V compared with the EEG cohort. We also demonstrated that despite the inaccuracy in adjusting the hippocampal AP coordinate when using MRI guidance, 36% of the hippocampal electrodes were recording in the dentate gyrus in the MRI cohort compared with 49% in the EEG cohort. This finding suggests that a 3D change in hippocampal orientation, which was less evident in the 2D MR images used in this study, compromised the estimation of the hippocampal AP coordinate. The 2D images were effective for determining the DV hippocampal target as the overall DV distribution of the electrode tip in the dentate gyrus was the same between sham and TBI rats.

One question is: Would presurgery MRIs help to improve the interpretation of postsurgery MRIs and the accuracy of electrode placements? In addition to the cost, it is important to note that unlike MRI, the histology-based rat brain atlas offers a benefit of using skull landmarks for calculation of coordinates for electrode positioning, that is, using the bregma as a reference point. Histological sections also give a higher spatial resolution and possibility to also assess the laminar placement of electrode tips.

Taken together, MR images were useful in targeting the AP and DV locations of perilesional intracortical electrodes in the chronic post-TBI phase. For hippocampal electrode implantations, the MR images were not as useful in targeting the AP location, but improved the precision of targeting the DV electrode tip into the dentate gyrus. To increase the accuracy of estimating the AP coordinate for intracerebral electrode implantation at the chronic post-TBI phase, we suggest using a 3D reconstruction of the brain to fully understand the effect of the ipsilateral cortical and hippocampal atrophy and orientation affecting the electrode placements. We propose the following MRI protocol for estimating the adjustments needed for atlas-based coordinates to successfully and accurately implant electrodes in the chronic post-TBI phase (see also Supplementary Figure S1):


#### *5.3. Chronic Intracerebral Electrode Implantation Did Not Affect the Cortical Lesion Area or Cortical and Hippocampal Atrophy over the 7-Month Follow-Up*

Chronic electrode implantation has been proposed to add tissue damage due to blood– brain barrier damage and chronic inflammation in the electrode path [40]. Therefore, we expected to see more cortical atrophy in the EEG than the MRI cohort. The cortical lesion areas were comparable in the EEG and MRI cohorts. Also, the percentage of the lesion area in cytoarchitectonic areas targeted by the electrode tips did not differ between the EEG cohort and MRI cohort. Taken together, the progression of the cortical lesion in the EEG cohort was not augmented by the chronically implanted electrodes.

We recently demonstrated that the hippocampus undergoes a series of post-TBI morphologic transformations, including atrophy and orientation changes due to neurodegeneration, white-matter atrophy, and expanding ventricles [23]. The present analysis

revealed a caudal shift of the hippocampus. Apparently, the expanding ventricles pushed the hippocampus backward and toward the midline.

#### **6. Conclusions**

Our study demonstrates the benefit of using MR images for adjusting atlas-based coordinates to improve the accuracy in chronic intracerebral electrode placements into atrophied and distorted brain areas, which is critical, e.g., for a high-quality recording of various epileptiform activities in candidate epileptogenic regions after TBI. Future studies should consider using T2-wt MRI slices much thinner than 800 μm or 3D spatial encoding with close to isotropic resolution to allow for accurate visualization in coronal, sagittal, and horizontal planes to enhance the accuracy of MRI-guided electrode implantations. Importantly, comparison of the cortical lesion area and cytoarchitectonic distribution between cohorts with 7-month- or 1-month-long electrode implantations did not reveal any worsening of brain atrophy by the chronic electrode implantation.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/biomedicines10092295/s1. Supplementary Table S1. The distribution of the dorsoventral location of the lower tip of the anterior and posterior intracortical electrodes in sham-operated and TBI rats. Supplementary Table S2. Location of the "virtual electrode". Summary of the locations of the anterior and posterior intracortical, and hippocampal electrodes in the MRI cohort, if implanted according to the atlas-based coordinates. Supplementary Figure S1. A schematic presentation of the MRI protocol for estimating the adjustments needed for atlas-based coordinates (See discussion for further details).

**Author Contributions:** Conceptualization, X.E.N.-E., O.G. A.P.; formal analysis, X.E.N.-E. and E.H.; funding acquisition, A.P. and O.G.; investigation, X.E.N.-E., R.I., E.H., E.M., P.A., R.C. and T.P.; methodology, X.E.N.-E., R.I. and O.G.; project administration, X.E.N.-E.; resources, O.G. and A.P.; software, P.A., R.C. and T.P.; supervision, O.G. and A.P.; visualization, X.E.N.-E. writing—original draft, X.E.N.-E. and A.P.; writing—review and editing, X.E.N.-E. and A.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was supported by the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement 602102 (EPITARGET)(AP), the Medical Research Council of the Academy of Finland (grants 272249, 273909, and 2285733-9) (AP), the National Institute of Neurological Disorders and Stroke (NINDS) Center without Walls of the National Institutes of Health (NIH) under award U54NS100064 (EpiBioS4Rx) (AP), and the Sigrid Jusélius Foundation (AP).

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki. All animal procedures were approved by the Animal Ethics Committee of the Provincial Government of Southern Finland and carried out in accordance with the guidelines of European Community Council Directives 2010/63/EU.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Acknowledgments:** We thank Jarmo Hartikainen and Merja Lukkari for their excellent technical help.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Review* **The Neuroinflammatory Role of Pericytes in Epilepsy**

**Gaku Yamanaka 1,\*, Fuyuko Takata 2, Yasufumi Kataoka 2, Kanako Kanou 1, Shinichiro Morichi 1, Shinya Dohgu <sup>2</sup> and Hisashi Kawashima <sup>1</sup>**


**Abstract:** Pericytes are a component of the blood–brain barrier (BBB) neurovascular unit, in which they play a crucial role in BBB integrity and are also implicated in neuroinflammation. The association between pericytes, BBB dysfunction, and the pathophysiology of epilepsy has been investigated, and links between epilepsy and pericytes have been identified. Here, we review current knowledge about the role of pericytes in epilepsy. Clinical evidence has shown an accumulation of pericytes with altered morphology in the cerebral vascular territories of patients with intractable epilepsy. In vitro, proinflammatory cytokines, including IL-1β, TNFα, and IL-6, cause morphological changes in human-derived pericytes, where IL-6 leads to cell damage. Experimental studies using epileptic animal models have shown that cerebrovascular pericytes undergo redistribution and remodeling, potentially contributing to BBB permeability. These series of pericyte-related modifications are promoted by proinflammatory cytokines, of which the most pronounced alterations are caused by IL-1β, a cytokine involved in the pathogenesis of epilepsy. Furthermore, the pericyte-glial scarring process in leaky capillaries was detected in the hippocampus during seizure progression. In addition, pericytes respond more sensitively to proinflammatory cytokines than microglia and can also activate microglia. Thus, pericytes may function as sensors of the inflammatory response. Finally, both in vitro and in vivo studies have highlighted the potential of pericytes as a therapeutic target for seizure disorders.

**Keywords:** pericytes; mural cells; cytokine; blood-brain barrier; neuroinflammation

#### **1. Introduction**

Accumulating evidence has demonstrated that the pathogenesis of epilepsy is linked to neuroinflammation and cerebrovascular dysfunction [1–6]. Traditionally, microglia had been considered to be responsible for the cytokine-centered immune response in the central nervous system (CNS); however, brain pericytes can respond to inflammatory signals, such as circulating cytokines, and convey this information to surrounding cells through chemokine and cytokine secretions [7–10]. Recent studies have demonstrated that pericytes may act as sensors for the inflammatory response in the CNS, as pericytes react intensely to proinflammatory cytokines when compared to other cell types (e.g., microglia) that constitute the CNS and factor-induced reactive pericytes can also activate microglia in vitro [9,11–13].

Pericytes provide physical support to the blood–brain barrier (BBB) and play an integral role in CNS homeostasis and BBB function [14]. Pericyte degeneration and/or dysfunction contribute to the loss of BBB integrity, which is an early hallmark of several neurodegenerative and inflammatory conditions [8,15,16]. Another notable feature of pericytes is their ability to regulate the migration of leukocytes across the brain microvascular endothelial cell (BMVEC) barrier, which secretes key molecules that support the

**Citation:** Yamanaka, G.; Takata, F.; Kataoka, Y.; Kanou, K.; Morichi, S.; Dohgu, S.; Kawashima, H. The Neuroinflammatory Role of Pericytes in Epilepsy. *Biomedicines* **2021**, *9*, 759. https://doi.org/10.3390/ biomedicines9070759

Academic Editor: Prosper N'Gouemo

Received: 31 May 2021 Accepted: 26 June 2021 Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

BBB barrier [17,18]. Recent research on the pathogenesis of epilepsy has begun to elucidate the mechanisms mediating peripheral-to-CNS cell infiltration in human and mouse models [19,20]. Pericytes may contribute to the mechanisms, while emerging research is investigating the extent of peripheral immune cell involvement in the inflammatory pathology of epilepsy.

The various functions of pericytes and their involvement in CNS diseases, including ischemic stroke [21], spinal cord injury [22], brain injury [23], and multiple sclerosis [24], has been reported.

The association between pericytes and epilepsy has attracted attention, while several recent studies have illustrated the contributions of pericytes to the pathogenesis of epilepsy [2,25–32]. These studies suggested that pericytes might participate in the pathogenesis of epilepsy, consisting of neuroinflammation and BBB damage and the interaction between peripheral and central immunity. Thus, evidence on the relationship between pericytes and the pathogenesis of epilepsy is gradually accumulating. Therefore, this study aimed to investigate the pathogenesis of epilepsy and pericytes because none of the review articles focused on this, even though therapeutic targets for pericytes in neurological disorders were investigated [17,33,34].

This review (1) explores the current literature regarding the role of pericytes in the pathogenesis of epilepsy and (2) highlights novel directions for research on therapeutic interventions for epilepsy that target pericytes. Given the paucity of knowledge on pericyte function in seizures and epilepsy-related pathologies, further studies are warranted to investigate pericytes as a potential therapeutic target for epilepsy treatment.

#### **2. What Are Pericytes?**

Pericytes were first described by the French scientist Charles-Marie Benjamin Rouget and were originally called Rouget cells in 1873 [35]. Later, this population was rediscovered by Zimmermann as a cell that shows a specific morphology around microvessels, and became widely known as a "pericyte" [36]. Pericytes are mural cells that are implanted in the basal membrane surrounding endothelial cells in capillaries and small vessels, including precapillary arterioles and postcapillary venules. Although the origin of all pericytes has not been clarified, blood vessels in the CNS are predominantly covered by neural crest cell-derived pericytes, while mesoderm-derived pericytes mainly contribute to blood vessel coverage in the trunk [37]. In the brain, pericytes constitute a vital component of the BBB/neurovascular unit (NVU) and cover the BMVECs lining the capillaries on the parenchymal side, where there are astrocytic end feet that enclose cerebral vessels, perivascular microglia/macrophages, and neurons [17,38,39]. Pericytes form a crucial component of the brain microvasculature and play an integral role in CNS homeostasis and BBB function [14] in normal physiological (Figure 1) and pathological conditions (Figure 2). A potential mechanism of pericyte action is the regulation of signaling through plateletderived growth factor receptor beta (PDGFRβ), which is commonly used as a marker of pericytes and regulates pericyte survival, proliferation, and migration signals [40]. In the CNS, platelet-derived growth factor-beta subunit (PDGF-BB) is released by endothelial cells and binds to PDGFRβ at the cell surface of pericytes to promote pericyte vascularization within the BBB [41]. The PDGFRβ signaling pathway is involved in pericyte survival and subsequent development as well as the function of the BBB during adulthood and senescence, as demonstrated by experiments in pericyte-deficient mice [17,38]. In addition to its role as a marker of CNS pericytes, PDGFRβ is expressed in oligodendrocyte precursor cell (OPC)/neuron-glial antigen 2 (NG2) parenchymal glial cells [2,25,42]. Other markers for pericytes exist (Table 1), but these remain inconclusive. Anatomical studies are required to investigate the characteristics of pericytes that possess longitudinal processes along vessels and contribute to BBB maintenance [15]. Pericytes in the brain are highly heterogeneous and have different morphologies as well as functions depending on their location in the vasculature [10]. Further, transgenic mice generated to study pericyte function may yield information on other cell types [43]. Therefore, "peripheral blood-specific" markers must

be used with caution [44]. Although there is no scientific consensus on what constitutes true pericytes [45], the current review focuses on studies using definitive pericyte-related markers and anatomy.

**Figure 1.** Regulatory functions of pericytes. In the central nervous system (CNS), platelet-derived growth factor-beta subunit (PDGF-BB) is released by endothelial cells and binds to PDGFRβ at the cell surface of pericytes to promote pericyte vascularization within the blood–brain barrier (BBB). Secretion of angiopoietin-1 (ANGPT-1) and plasminogen activator inhibitor type 1 (PAI-1) from pericytes promotes the development of vascular endothelial cells and contributes to the maintenance of the BBB (1). Pericytes maintain neuronal health by secreting factors such as nerve growth factor (NGF), brain-derived nerve growth factor (BDNF), and pleiotrophin (2). Pericytes are involved in angiogenesis by secreting ANGPT-1 and erythropoietin (3) and produce a factor (Lama2) that facilitates the differentiation of oligodendrocyte progenitor cells (OPCs) into mature oligodendrocytes (4).

**Figure 2.** In pathological conditions, pericytes generate various inflammatory factors. Pericytes secrete IL-6 that can polarize parenchymal microglia to a proinflammatory phenotype to activate microglia (1). The secretion of chemokines (CCL2, CXCL1, CXCL8, and CXCL10) by pericytes recruits leukocytes to the CNS parenchyma via the upregulation of ICAM-1 and VCAM-1 adhesion molecules on the endothelium (2). MMP-9 secretion stimulates the production and secretion of vascular endothelial growth factor (VEGF), resulting in endothelial dysfunction (3). Secretion of reactive oxygen species/reactive nitrogen species (ROS/RNS), nitric oxide (NO), and prostaglandins (PGE2) by pericytes lead to vasodilation and breaching of the blood–brain barrier. Pericytes themselves are morphologically altered by inflammatory mediators (4).


**Table 1.** Common markers used to identify pericytes in the central nervous system of mice that also label other cell types.

Note: NSCs, neural stem cells; OPCs, oligodendrocyte progenitor cells; SMCs, smooth muscle cells.

#### **3. Pericytes and Neuroinflammation**

Evidence accumulated from experimental models and human samples implicates immunological processes in the pathogenesis of epilepsy [1,4]. The involvement of pericytes in the CNS immune responses has attracted significant attention. Pericytes present heterogeneous signals to the surrounding cells and actively modulate inflammatory responses in a tissue- and context-dependent manner. The expression of various pattern-recognition receptors (PRRs), including toll-like receptors (TLRs) and nucleotide-binding and oligomerization domain (NOD)-like receptor families, has been detected in brain pericytes [52]. Given the abundance of surface receptors, pericytes can respond to inflammatory mediators, such as monocyte chemoattractant protein-1 (MCP-1/CCL2) and tumor necrosis factor (TNF)-α, which in turn induce the secretion of CCL2, nitric oxide (NO), and several cytokines [7–9,53]. Pericytes act as promoters of both the innate and adaptive immune system [43]. In the CNS, microglia are a hallmark of the immune response, which produce cytokines such as interleukin (IL)-1β, TNF-α, IL-6, and various other chemokines [54], and related effector pathways, including cyclooxygenase-2 (COX-2)/prostaglandin (PGE2) and complement factors [55]. The rapid activation of microglia impairs neuronal function by inducing inflammatory mediators, such as NO, reactive oxygen species (ROS), and proinflammatory cytokines [56,57].

Pericytes have been shown to be more sensitive to proinflammatory cytokines compared to other cells in the NVU [9,11–13]. Specifically, cytokine and chemokine release profiles from brain pericytes in response to TNF-α are distinct to those of other cell types comprising the NVU, and TNF-α-stimulated pericytes release macrophage inflammatory protein (MIP)-1α and IL-6. Among BBB cells, pericytes stimulated with TNF-α induced the highest levels of *iNOS* and IL-1β mRNA expression, which indicates the activation of BV-2 microglia [9]. The mechanism underlying TNF-α-induced IL-6 release involves the inhibitor kappa B (IκB)-nuclear factor kappa-light-chain-enhancer of activated B cells (NFκB) and the Janus family of tyrosine kinase (JAK)-signal transducer and activator of transcription (STAT) 3 pathways [13]. NFκB plays a key role in inflammation, immune, and stress-related responses, as well as in the regulation of cell survival and in the growth of neural processes in developing peripheral and central neurons [58]. These findings indicate that the activated brain pericytes trigger the development of uncoordinated NVU function, including glial activation, and may act as sensors at the BBB in TNF-α-mediated brain inflammation.

Pericytes also release anti-inflammatory factors, highlighting their involvement in regeneration and protection [7,59,60]. Pericytes respond to lipopolysaccharide (LPS), secrete anti-inflammatory cytokines such as IL-10 and IL-13 [61], and produce neurotrophins such as nerve growth factor (NGF) and brain-derived neurotrophic factor (BDNF), which regulate neuronal development [42,62]. Pericytes upregulate neurotrophin-3 production in response to hypoxia, resulting in increased NGF production in astrocytes, thereby protecting neurons from hypoxia-induced apoptosis [62]. These actions highlight the neuroprotective functions of pericytes under pathological conditions.

#### **4. Pericytes and Epilepsy**

Table 2 summarizes the research on pericytes and epilepsy.




**Table 2.** *Cont*.

CCI, controlled cortical impact; FCD, focal cortical dysplasia, HS, hippocampal sclerosis; IP, intraperitoneal; KA, kainic acid; PDGF-BB, platelet-derived growth factor-beta subunit; SE, status epilepticus; TBI, traumatic brain injury; TLE, temporal lobe epilepsy.

#### **5. Blood-Brain Barrier Disruption in the Pathogenesis of Epilepsy**

Experimental evidence of BBB impairment in the pathogenesis of epilepsy has been demonstrated in patients and animal models [64–67], which is a hallmark of epilepsy. BBB disruption can also directly induce seizure activity and exacerbate epileptogenesis; the relationship between epilepsy and BBB breakdown is bidirectional [64,65].

BBB dysfunction and subsequent infiltration of serum albumin into the brain leads to changes in epileptogenesis, including astrocyte changes, neuroinflammation, excitatory synapse formation, and pathological plasticity [68,69]. These BBB alterations are not only due to leakage, as demonstrated by Evans Blue staining [65]. There is involvement of various inflammatory mediators as nondisruptive changes at the molecular level of pericytes are also involved in the changes of the BBB; specifically, they secrete various mediators as

follows: IL-1β, TNF-α, IFN-γ, matrix metalloproteinases (MMPs), ROS/reactive nitrogen species (RNS), (NO), and prostaglandin E2 (PGE2). Pericyte-derived MMP-9 upregulation in the cerebral microvasculature can cause endothelial dysfunction through degradation of tight junctions and extracellular matrices, resulting in subsequent pericyte loss from the microvasculature and BBB disruption [11,43]. Moreover, the secretion of ROS/RNS, NO, and PGE2 lead to vasodilation and breaching of the BBB [9]. Epileptic seizures can cause pericytes surrounding the blood vessels to rearrange [2] and morphologically alter, which is facilitated by the inflammatory mediators [29,30]. These series of alterations are thought to be linked to the pathogenesis of epilepsy, although further details are warranted.

#### **6. Leukocyte Recruitment and Peripheral-to-Central Infiltration**

Pericytes regulate the migration of leukocytes across the BMVEC barrier and secrete key molecules that support the BBB [17,18]. Chemokines (CCL2, CXCL1, CXCL8, and CXCL10) secreted by pericytes in both basal and inflammatory states recruit peripheral immune cells, including monocytes, B and T cells, and neutrophils, to the CNS parenchyma via upregulation of intercellular adhesion molecule-1 (ICAM-1) and vascular cell adhesion molecule-1 (VCAM-1) on the endothelium [7–9,70]. Although the human brain is considered an immune-privileged area [68,71], this is not preserved during inflammatory conditions. Analysis of brain parenchyma in patients with epilepsy showed that there have been both positive [72,73] and negative [74] reports on the occurrence of infiltration of peripheral leukocytes into the brain tissue. Recent experimental research demonstrated that peripheral-to-CNS cell infiltration, particularly monocytes, occurs in the status epilepticus (SE) model, without evidence of infections or immune disorders [20,75,76]. The possibility of classifying peripheral monocytes and indigenous microglia, which have been considered difficult to differentiate, has been increased using genetic engineering [75,77,78].

In chemokine receptor 2 (CCR2)-knockout mice, the CCL2 receptor, which blocks peripheral monocyte invasion into the brain tissue, attenuated neuronal damage in SE models [75]. Analysis of the brain tissue from pediatric patients with drug-resistant epilepsy (DRE) revealed that seizure frequency was correlated with the number of infiltrating peripherally activated CD3+ T cells and monocytes, but not microglia [19]. Current analysis of pediatric patients with DRE also demonstrated a correlation between the number of seizures and intracellular IL-1β levels in monocytes [79], while experimental data and human research attributed seizure-induced neuronal death to the activation of resident microglia [78,80]. Whether the peripheral monocytes or the resident microglia are the primary triggers of epilepsy, as well as the extent to which the infiltrated cells are significant, remains to be determined; nevertheless, the combination of the roles of the pericytes in maintaining the BBB integrity, producing inflammatory mediators, and recruiting leukocytes indicate that the pericytes could be intimately involved in the pathogenesis of epilepsy.

#### **7. Clinical Evidence Links Pericytes to Epilepsy**

The disarray of the pericyte-basal lamina interface in patients with epilepsy was first described in 1990 [63]. Evidence of pericyte degeneration with basement membrane unit thickness and cytoplasmic density has also been reported in most of the spiking area microvessels in human brain tissues of intractable complex partial seizures using an electron microscope [63].

With the advent of PDGFRβ, though a nonspecific CNS pericyte marker, the immunostaining reports of the presence of PDGFRβ+ cells have emerged in the brain specimens of patients with intractable epilepsy in focal cortical dysplasia (FCD) and temporal lobe seizures (TLE) [2,25,29]. In tissues from patients with refractory TLE and hippocampal sclerosis (HS), the presence of PDGFRβ+ cells associated with blood vessels and parenchyma was observed, although findings were heterogenous [2]. Indeed, the highest perivascular PDGFRβ immunoreactivity was detected in patients with TLE-HS, specifically in the microvasculature [2]. Tissue from patients with cryptogenic epilepsy has exhibited a similar immune response pattern, although to a lesser extent than that of FCD. Increased perivascular PDGFRβ immunoreactivity was associated with increased hippocampal vascularization in the cells of patients with TLE-HS [25].

Another study of TLE and FCD specimens revealed robust PDGFRβ-positive cell pericyte immunoreactivity surrounding the blood vessels, particularly in TLE with HS specimens, with aggregation of IBA1/HLA microglial cells and pericyte-microglia outlining the capillary wall [29]. The morphological changes in pericytes were induced by proinflammatory cytokines, including IL-1β, TNFα, and IL-6; in particular, IL-6 exposure was drastically associated with apoptosis, suggesting pericyte damage [29].

Collectively, the accumulation of pericytes (PDGFRβ-positive cells) in the cerebral vascular regions was consistently observed in patients with refractory epilepsy [2,25,29]. The degree of accumulation correlates to some extent with the clinical picture [25,29], and morphological changes of the pericytes might be due to proinflammatory cytokines [29]. In addition, the amount of angiogenesis, which is associated with epileptogenesis, was related to the number of PDGFRβ-positive cells [25], suggesting a relationship between PDGFRβ-positive cells and the pathogenesis of epilepsy.

#### **8. Experimental Evidence Links Pericytes to Epilepsy**

An in vivo study of NG2DsRed mice, which enabled the visualization of cerebrovascular pericytes, revealed heterogeneous perivascular prominence of NG2DsRed cells with PDGFRβ expression in an SE model induced by intraperitoneal kainic acid (KA) [2]. These heterogeneous perivascular patterns of PDGFRβ+ cells are inconsistent with the aforementioned human tissue findings [2,25,29], which have also been observed in a rat model of neurovascular dysplasia SE, particularly in the hippocampus with a neurovascular dysplasia SE rat model [25].

An in vitro and in vivo study by Milesi et al. demonstrated that the parenchymal and vascular PDGFRβ+ cells were redistributed, alongside partial colocalization of vascular and parenchymal PDGFRβ+ cells with NG2DsRed and NG2, but not with IBA-1 [2]. These findings, suggesting that the accumulation of pericytes and microglia is associated with epileptic seizure events, have been documented in recent studies [29,30].

Klement et al. employed a model of TLE (associated with HS) in NG2DsRed mice to assess the impact of seizure progression on capillary pericytes and surrounding glial cells [29]. In vivo, SE mice presenting with spontaneous recurrent seizures (SRS) exhibited disorganized NG2DsRed-positive pericyte somata in the hippocampus at 72 h and 1 week after SE (epileptogenesis) in the hippocampus. Pericyte modifications clustered with IBA1-positive microglia, surrounding capillaries, and overlapped topographically with pericytes lodged within microglial cells [29]. Residual microglial clustering was also observed surrounding NG2DsRed pericytes in SRS, proinflammatory mediators, such as IL-1β, IL-6, TNF-α, and particularly IL-1β; however, the in vitro study in humans revealed that IL-6 induced these morphological changes of pericyte-microglia clustering in NG2DsRed hippocampal slices [29]. In addition, Klement et al. also reported a pericyte-glia perivascular scar with capillary leaks in the hippocampus during seizure activity. These scars in the cornu ammonis region developed an abnormal distribution or accumulation of extracellular matrix collagen III/IV as the seizure progressed [30]. In vitro experiments induced by 4-aminopyridine and low-Mg2+ conditions repeated seizures that cause vasoconstriction associated with the depolarization of mitochondria in pericytes and gradual neurovascular disconnection, suggesting that the pericyte damage causes vascular dysfunction in epilepsy [31]. The gradual progression of neurovascular decoupling during recurrent seizures suggests that pericyte damage induces vascular dysfunction in epilepsy (Figure 3) [31].

**Figure 3.** Schematic representation of the events linking pericytes to epilepsy. Status epilepticus leads to redistribution and remodeling of cerebrovascular pericytes, potentially contributing to blood–brain barrier permeability [2,28,29]. A significant clustering of microglia/macrophages around pericytes occurs one week after the attack, although pericyte proliferation is significantly increased as early as 72 h [29]. These series of pericyte-related modifications are promoted by proinflammatory cytokines, including IL-1β, TNFα, and IL-6. Alterations caused by IL-1β, which is one of the cytokines most deeply involved in the pathogenesis of epilepsy, were most pronounced. These pericyte-associated modifications and pericyte-microglia clustering may be facilitated by IL-1β [29], and pericyte-glial scarring with collagens III and IV process leaky capillaries during seizure progression [30]. Recurrent seizures can lead to pericytic injury with neurovascular decoupling and BBB dysfunction at the arterial and capillary levels. Moreover, capillary vasoconstriction is accompanied by a loss of mitochondrial integrity in pericytes [81]. In vitro and in vivo studies have highlighted the potential of pericytes as a therapeutic target for seizure disorders [28,30,32].

#### **9. Prospects for Pericyte-Mediated Epilepsy Therapy**

PDGFRβ can regulate pericyte survival, proliferation, and migration signals and is commonly used as a marker for pericytes [40]; PDGFRβ suppression has been proposed as a possible treatment for epilepsy [28,30,32].

As described above, a pericyte-glia perivascular scar with capillary leaks induced by seizures and a high expression of PDGFRβ transcript and protein levels were detected [30]. In the organotypic hippocampal cultures, PDGFRβ reactivity surrounding capillaries is also enhanced by electrographic activity and was reduced by PDGF-BB (a PDGFRβ agonist) and PDGFβ inhibitor imatinib [30]. Furthermore, PDGF-BB can reduce mural cell loss, vascular pathology, and epileptiform electroencephalography activity in a KA-induced SE model [28]. Recently, traumatic brain injury (TBI) has been highlighted as a major factor in epilepsy owing to certain intractable cases. The evaluation of the involvement of pericytes in the pathogenesis of epilepsy was performed using a controlled cortical impact (CCI) device. PDGFRβ levels were significantly increased following CCI in the injured ipsilateral hippocampus; pilocarpine-induced seizures can be regulated by imatinib treatment in this CCI model [32]. The efficacy of imatinib was also observed in vitro.

The findings from both in vitro and in vivo studies highlight the potential of pericytes as a therapeutic target for seizure disorders, as indicated by the efficacy of PDGF-BB and imatinib in blocking PDGFRβ. However, both PDGFRβ and PDGF-BB are required for the pericyte coating of the BBB in the developing CNS [38,41]. Under pathological conditions, mural cells in the immediate postacute phase (SE, ischemic stroke, and head trauma) require support from the PDGFRβ activation [28]; hence, the inflammatory involvement of PDGFRβ may be relevant in long-term progression as well as in chronic stages.

When considering the pharmacological modulation of pericyte signaling pathways as a means of attenuating disease progression and capillary pathology, the impact of pericyte modulation in the epileptic brain must consider the activation state of the glial cells and the disease stage (e.g., acute vs. chronic) [29]. Further, considering the distinct functions of PDGFRβ at different developmental stages, the timing of PDGFRβ inhibition needs to be carefully studied; moreover, avoiding imatinib in the acute phase of the disease may be considered. It remains debatable whether the changes in pericytes and accumulation of microglia associated with PDGFR expression in this series of studies should be suppressed.

Transforming growth factor-beta 1 (TGFβ1) is a multifaceted cytokine in the brain that plays a role in regulating cell proliferation, differentiation, survival, and scar formation [82,83]. Since 1989, the possibility of PDGF-induced TGF-β signaling has been suggested [84]; PDGFR-β and TGF-β with PDGFR-β might mediate the endothelial cell/pericyte interaction to protect the BBB integrity [33]. The potential involvement of TGF-β in epileptogenesis has been recognized from an experimental model showing TGF-β upregulation as part of the inflammatory response [85]. Microarray analysis of TGFβ1-stimulated human brain pericytes isolated from intractable TLE demonstrated inhibition of pericyte proliferation and phagocytosis by TGFβ1 [27]. However, TGFβ1 also enhanced the expression of IL-6, MMP-2, and NOX4, which can disrupt BBB functioning; thus, these reactions caused by TGFβ1 might not lead to the treatment of the neurovascular system [27].

Although the brain pericyte-derived TGF-β contributes to the upregulation of BBB functions [86], suppression of TGFβ1 indicates improvement in epilepsy [87]. Losartan, an angiotensin-type 1 receptor (AT1) antagonist, prevents phosphorylation of Smad proteins of TGF-β signaling [88,89], which has demonstrated both neuroprotective and antineuroinflammatory effects [90–92].

These in vitro studies also suggest that human-derived pericytes are morphologically altered by proinflammatory cytokines that induce apoptosis [29], indicating the potential of targeting IFN-γ for pericyte-mediated epilepsy treatment [26]. IFN-γ is a central component of the CNS inflammatory response and is secreted by microglia, astrocytes, endothelial cells, and circulating immune cells [93–95]. This classical inflammatory mediator has been implicated in CNS diseases, including epilepsy [96,97]. Altering the proportion of microglial phenotypes via IFN-γ treatment improved the prognosis in a mouse model of epilepsy [98].

Notably, in epileptiform conditions, IL-1β, a neurotoxic cytokine and one of the cytokines chiefly involved in the pathogenesis of epilepsy, prominently contributes to the morphological changes in the pericytes [29]. There is evidence that the IL-1/IL-1R1 axis plays an important role in the inflammatory response in epilepsy, as presented by Vezzani et al. in an excellent review [4,99]. IL-1β agonist, the IL-1 receptor antagonist (IL-1RA), has already been tested for clinical application for epileptic syndromes using anakinra, and has shown favorable clinical outcomes [100–103]. The use of anakinra on pericytes in status epilepticus has not yet been investigated. To ensure the involvement of pericytes in epilepsy, it is worthwhile to confirm that anakinra suppresses the morphological changes in pericytes and reduces seizures.

Previous reports have demonstrated that inhibition of pericytes could have positive effects of neuroprotection [26,28,30,32]; however, there is also a concern that the suppression of pericytes by TGFβ1 may not necessarily have a positive effect on the CNS [27]. Since TGFβ1 suppresses pericyte phagocytosis and reduces the expression of central leukocyte trafficking chemokines and adhesion molecules while increasing the expression of proinflammatory cytokines and enzymes that promote BBB disruption, a paradoxical reaction has been reported [27]. The TGFβ1 response of pericytes may differ from the

anti-inflammatory response of microglia [104–107]; therefore, further studies are required to obtain any effect on this nonuniform response.

In the pathogenesis of epilepsy, pericytes adopt a phenotype that is neither solely pro- nor anti-inflammatory [27]. Merely suppressing pericytes may not be sufficient to improve the treatment of epilepsy, and it may be necessary to seek a treatment tailored to the affected child in combination with various therapies that have been introduced in recent reviews [108].

#### **10. Conclusions**

In this review, we present evidence for the substantive role of pericytes in the pathogenesis of epilepsy. The roles of pericytes in maintaining BBB integrity, producing inflammatory secretions, and recruiting leukocytes highlights the potential role of pericytes in the pathogenesis of epilepsy. Pericytes may also act as sensors of inflammatory processes in the CNS and regulating them may lead to the development of novel therapies for epilepsy. However, as there remains a lack of absolute molecular markers for pericytes, and since pericytes originate from multiple cellular sources and vary in morphology, localization as well as function in different tissues leaves several issues to be addressed. In addition, we are unable to determine whether brain inflammation is an initiator or a consequence of a systemic inflammatory process.

Several reports have suggested entry points that may also act as a basis for various neurovascular therapies, including anakinra [100,101] and losartan [87], though the level of evidence for both drugs is limited for the establishment of treatment for epilepsy. These drugs provide an avenue for novel therapeutic, anti-inflammatory, or cerebrovascular repair to mitigate epileptic pathophysiology. Unfortunately, definitive treatments for epilepsy are currently lacking. BBB integrity and systemic peripheral inflammation may contribute to epilepsy and hold potential for molecular biomarkers and targets in the treatment of epilepsy. Moreover, human pluripotent stem cell-derived brain pericyte-like cells induced BBB properties in BMECs, resulting in strengthening of the barrier and a reduction in transcytosis [109]. These stem cell techniques could be applied to examine the possibility of new strategies to selectively target pericytes and the role of pericytes in epilepsy more specifically. Novel tools to control pericytes should be developed to target inflammatory vascular-related processes during seizure progression or activity.

**Author Contributions:** Conceptualization, G.Y.; investigation, K.K. and S.M.; writing—original draft preparation, G.Y.; writing—review and editing, F.T.; visualization, S.D.; supervision, Y.K. and H.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was funded by the Kawano Masanori Memorial Foundation for Promotion of Pediatrics in Japan under grant number 30-7 and the Japan Epilepsy Research Foundation under grant number 20012. APC funded by the Japan Epilepsy Research Foundation.

**Institutional Review Board Statement:** We confirm that we have read the journal's position on the issues associated with ethical publication and affirm that this report is consistent with these guidelines.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The datasets generated and/or analyzed during the current study are available at the PubMed database repository (https://pubmed.ncbi.nlm.nih.gov/, accessed on 31 May 2021).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Review* **Targeting the Ghrelin Receptor as a Novel Therapeutic Option for Epilepsy**

**An Buckinx 1, Dimitri De Bundel 1, Ron Kooijman <sup>2</sup> and Ilse Smolders 1,\***


**Abstract:** Epilepsy is a neurological disease affecting more than 50 million individuals worldwide. Notwithstanding the availability of a broad array of antiseizure drugs (ASDs), 30% of patients suffer from pharmacoresistant epilepsy. This highlights the urgent need for novel therapeutic options, preferably with an emphasis on new targets, since "me too" drugs have been shown to be of no avail. One of the appealing novel targets for ASDs is the ghrelin receptor (ghrelin-R). In epilepsy patients, alterations in the plasma levels of its endogenous ligand, ghrelin, have been described, and various ghrelin-R ligands are anticonvulsant in preclinical seizure and epilepsy models. Up until now, the exact mechanism-of-action of ghrelin-R-mediated anticonvulsant effects has remained poorly understood and is further complicated by multiple downstream signaling pathways and the heteromerization properties of the receptor. This review compiles current knowledge, and discusses the potential mechanisms-of-action of the anticonvulsant effects mediated by the ghrelin-R.

**Keywords:** epilepsy; ghrelin; ghrelin receptor

#### **1. Introduction**

Epilepsy is a neurological disease characterized by spontaneous and recurrent seizures [1]. With approximately 50 million patients, it is one of the most common neurological diseases worldwide [2]. Despite the availability of a wide range of antiseizure drugs (ASDs), up to 30% of patients suffer from pharmacoresistant epilepsy [2], of which a large proportion has temporal lobe epilepsy (TLE) [3,4]. This highlights the urgent need for the development of novel pharmacological treatment options.

One of these potential options is the orexigenic peptide, ghrelin. Ghrelin exerts both peripheral as well as central effects, and is primarily secreted by X/A-like cells in the stomach [5], but also, to less extents, in the small intestine, kidney, testis, pancreas, and the brain [5–10]. Peripherally, ghrelin plays an important role in gastric acid secretion, gastric emptying, and gastric motility [11–13], and it maintains glucose homeostasis via the inhibition of the insulin response to glucose administration [14]. Additionally, ghrelin is generally accepted to be a cardioprotective peptide [15,16].

In the central nervous system (CNS), ghrelin and its receptor are best known for their critical role in food intake, mediated by neuropeptide Y and agouti-related peptide [17–19] (reviewed in [20]). Additionally, ghrelin confers a regulatory role on growth hormone (GH) release [19], is implicated in learning and memory [21–23], modulates motivation and reward [24,25], and regulates the stress response (reviewed in [26]).

Soon after its discovery in 1999 [27], the interest in ghrelin within the context of epilepsy started to emerge. Ghrelin levels were shown to be altered in epilepsy patients, and ghrelin administration in preclinical seizure and epilepsy models is considered to be

**Citation:** Buckinx, A.; De Bundel, D.; Kooijman, R.; Smolders, I. Targeting the Ghrelin Receptor as a Novel Therapeutic Option for Epilepsy. *Biomedicines* **2022**, *10*, 53. https:// doi.org/10.3390/biomedicines10010053

Academic Editor: Prosper N'Gouemo

Received: 30 November 2021 Accepted: 24 December 2021 Published: 27 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

anticonvulsive [28–31]. However, up until now, the exact mechanism-of-action remains to be understood.

#### **2. Ghrelin and Its Receptor**

The main molecular form of ghrelin is a 28-amino acid (AA)-long peptide, with the active form containing a unique acylation on serine 3 [32]. Ghrelin is transcribed as a 117-AA-long preproghrelin. This is cleaved to render proghrelin, after which it undergoes acylation on serine 3, established by the membrane-bound enzyme, ghrelin O-acyltransferase (GOAT), which is distributed in a similar manner to ghrelin [33,34]. The acylation is either an octanoylation (eight-carbon fatty acid) or decanoylation (ten-carbon fatty acid) [33,34]. This action is followed by further processing of the 94-AA-long acylated pro-ghrelin by prohormone convertases 1/3 (PC1/3), which results in acylated ghrelin (AG), and also yields the mature peptide, obestatin [35]. Acylation on serine 3 was first believed to be imperative for the ability of ghrelin to bind to its receptor and to exert ghrelin's biological function [27]. Later, it became clear that desacyl ghrelin (DAG) is not completely devoid of physiological actions, as it was shown to also induce food intake, albeit through orexin neurons and not ghrelin receptor (ghrelin-R)-expressing neurons [36]. On the other hand, the anticonvulsant effects elicited by DAG required the presence of the ghrelin-R [37]. DAG shares some physiological functions similar to ghrelin but antagonizes others. Therefore, opposite effects might be mediated via a distinct receptor, and similar effects may be mediated by the ghrelin-R [38].

In human plasma, circulating esterases deacylate ghrelin, and 90% of total ghrelin consists of DAG, while only 10% consists of acylated ghrelin [39]. Ghrelin is rapidly cleared from plasma, with a plasma half-life ranging from 9–13 min for ghrelin, and 27–34 min for total ghrelin, including DAG [40,41]. Although the plasma concentration of DAG is much higher than that of ghrelin, its binding capacity to the ghrelin-R is substantially lower compared to ghrelin [38,42], which may explain why DAG was initially considered to be the "nonactive" variant of the peptide.

Recently, GOAT was shown to be expressed on the cell surface of mature bone marrow adipocytes, and to be necessary for DAG to promote adipogenesis in mice [43]. In line with this observation, GOAT was shown to be localized in the hilar border of dentate gyrus (DG) in the hippocampi of mice, and the incubation of live hippocampal slice cultures with DAG showed equal binding to the ghrelin-R as incubation with ghrelin, reliant on both the ghrelin-R and GOAT expression [44]. These data suggest that the local reacylation of DAG via GOAT expression at the cell surface may occur, and that it may be relevant for the biological functions of DAG mediated via the ghrelin-R.

#### *2.1. Signaling Pathways and Heteromerization Complicate Ghrelin-R Signaling*

Ghrelin establishes its numerous effects by interacting with its G-protein-coupled receptor (GPCR), of which two isoforms exist: the full-length (366 amino acids (AA) long) 7-transmembrane GPCR GHSR1a (a growth hormone secretagogue receptor, denoted as "ghrelin-R"), and a shorter (289 AA long) 3 -truncated variant, GHSR1b [27,45]. This nonsignaling short variant lacks the ability to exert biological effects in response to ghrelin and hampers the cell surface expression of the functional GHSR-1a variant, thus acting as a coregulator of ghrelin-R signaling [46,47].

The ghrelin-R is present in the brain and the periphery. The peripheral sites of ghrelin-R expression include the pancreas, spleen, bone tissue, cardiac tissue, the thyroid gland and immune cells, the adrenal glands, adipose tissue, and the vagal afferents [45,48].

Centrally, the ghrelin-R is widely expressed in a variety of brain areas and shows high expression levels in several nuclei of the hypothalamus, among which are the arcuate nucleus and the anterior hypothalamic nucleus. The receptor is further expressed in the olfactory bulb, the neocortex, in a variety of nuclei in the midbrain, in the pons, and in the medulla oblongata. These include the globus pallidus, the area postrema, the nucleus tractus solitarius, the substantia nigra, and the ventral tegmental area. In the hippocampus, the ghrelin-R was shown to be modestly expressed in the Cornu Ammonis (CA)1, compared to the higher expression levels in CA2, CA3, and DG of the mouse brain [45,49–51].

The expression of the ghrelin-R is highly dynamic, and depends on the developmental stage [52], the disease states [53,54], the metabolic state of the organism [55], or ghrelin availability [49]. Additionally, the receptor has a rich molecular pharmacology, with a multitude of signaling pathways associated with the receptor, and an ability to alter canonical ghrelin-R signaling via the formation of functional heteromeric complexes with other receptors. These factors contribute to diverse ghrelin-R signaling patterns.

The signaling pathways downstream of the ghrelin-R include Gαq/11, Gαi/o, and Gα12/13 signaling, followed by β-arrestin recruitment [56–58]. The canonical Gα<sup>q</sup> protein activates the phospholipase C (PLC)—inositol 1,4,5-triphosphate (IP3)—diacylglycerol (DAG) pathway, which leads to an intracellular calcium (Ca2+) increase [59,60]. Gαi/o activates phosphatidylinositol-3-kinase (PI3K) [61] and reduces cyclic AMP (cAMP) levels via reduced adenylyl cyclase (AC) activity [60]. Gα12/13 signaling is associated with the activation of Ras homolog family member A (RhoA), other Rho guanine exchange factors, and their associated Rho kinases [62] (reviewed by [63]). Finally, G-protein-mediated signaling is halted via the recruitment of β-arrestin to the receptor [64], which not only vouches for the desensitization and internalization, but may also activate G-protein independent signaling pathways (reviewed in [65,66]). Recent studies have shown the ability of the β-arrestin-mediated activation of ERK1/2, mitogen-activated protein kinase (MAPK), the Akt/protein kinase B (PKB) pathways, and RhoA signaling [56,61,67] (Figure 1).

**Figure 1.** Signaling pathways associated with the ghrelin receptor. The ghrelin receptor employs Gαq/11 signaling, Gαi/o signaling, and Gα12/13 signaling, followed by β-arrestin recruitment. Each G-protein/β-arrestin is associated with physiological effects. AC: adenylyl cyclase; AMPK: adenosinemonophosphate-activated protein kinase; ERK: extracellular signal-regulated kinase; mTOR: mammalian target of rapamycin; p-: phosphorylated-; RhoA: Ras homolog family member A; SRE: serum response element. Created with BioRender.com.

The ghrelin-R confers extraordinarily high intracellular signaling in the absence of ghrelin or a ghrelin-R full agonist, signaling at approximately 50% of its maximal capacity [67,68]. Constitutive activity includes signaling via G-proteins, while it does not entail β-arrestin-mediated endocytosis [67–69].

#### *2.2. What Is Known about Ghrelin's Central Availability?*

Ghrelin, DAG, or synthetic compounds must access the brain to be centrally active. The transport of ghrelin across the blood–brain barrier (BBB) has been shown to occur via saturable mechanisms in mice [23,70], and does not depend on the expression of the ghrelin-R [71]. The notion that the BBB may be compromised in epilepsy should be taken into account, which would facilitate the availability of ghrelin to the CNS.

Additionally, systemically injected ghrelin was shown to cross the fenestrated capillaries in the circumventricular organs (CVO) via passive diffusion, and dose-dependently impacted more distant brain areas [72]. Finally, fluorescent ghrelin was shown to internalize in ependymal cells located in the choroid plexus and in β-type tanycytes, which constitute the foundation of the blood–cerebrospinal fluid (CSF) barrier (BCSFB). Fluorescent ghrelin was detected in periventricular hypothalamic tissue, and decreased with distance from the third ventricle [73]. The transport of ghrelin via the BCSFB depends predominantly on the presence of the ghrelin-R [74]. The kinetics of diffusion into the brain via the BCSFB is somewhat slower compared to the diffusion via the CVOs, with the CSF ghrelin concentrations peaking approximately 30 min after the ghrelin plasma concentration peak, depending on plasma ghrelin levels [75]. Additionally, an in vitro study showed that ghrelin was internalized in rat primary tanycytes via clathrin-coated vesicles [76].

Up until now, it has remained incompletely understood whether ghrelin is centrally available in areas more remote from the aforementioned barriers. It is possible that circulating ghrelin reaches certain permeable parts of the brain, and affects other areas indirectly, via the innervation of the nuclei located in the vicinity of accessible brain parts. This was shown in the case of the area postrema, which directly notes alterations in plasma ghrelin levels and innervates the nucleus tractus solitarius [72]. Additionally, central ghrelin expression may serve as an explanation for the high ghrelin-R expression in brain areas that are seemingly inaccessible to circulating ghrelin [9,27]. Indeed, central ghrelin messenger ribonucleic acid (mRNA) expression and immunoreactivity have been shown in multiple studies; however, there are also some studies refuting this notion (reviewed in [77]).

#### **3. Studies in Humans**

In humans, the majority of total circulating ghrelin consists of DAG, due to the deacylation of AG [39]. The acylation is located at the N-terminal part of the peptide, while the rest of the molecule is equivalent between AG and DAG. The studies outlined below do not always specify the portion of the peptide that is recognized by the used assays, nor is this information always available on the manufacturer's website. Failing to specify which isoform was measured may explain some of the observed interstudy variations. Most of the studies investigating AG or DAG levels assessed this peptide in plasma (in a small number of studies, the saliva and urine ghrelin levels were also assessed) after overnight fasting and were conducted in children and adolescents. Ghrelin levels are negatively correlated with age [78], and even pubertal children have significantly lower total plasma ghrelin levels compared to prepubertal children [79]. Therefore, in this review, a distinction is made between studies on adults and studies on children.

#### *3.1. Adults*

Up to now, there has been no general consensus regarding the differences in interictal ghrelin levels between adult epilepsy patients and healthy subjects. Three studies showed lower ghrelin levels in seizure-controlled epilepsy patients compared to healthy controls [80–82], while two studies did not detect differences in plasma ghrelin levels between epilepsy patients and controls [83,84]. One study demonstrated that patients with seizure-controlled epilepsy had significantly higher serum ghrelin compared to healthy controls [85]. Three studies demonstrated that patients suffering from focal epilepsy had higher ghrelin plasma levels compared to patients suffering from generalized seizures [80,81,85]. Two studies were not able to replicate this finding [84,86] (Table 1).

**Table 1.** Overview of interictal ghrelin levels in adults with focal and generalized epilepsy. AG: acyl ghrelin; ASD: antiseizure drug; CBZ: carbamazepine; DAG: desacyl ghrelin; DR-TLE: drug-resistant temporal lobe epilepsy; PHT: phenytoin; Ref: reference; TLE: temporal lobe epilepsy; VPA: valproic acid. \* the different ghrelin levels in epilepsy patients versus controls. \*\* the differences in generalized epilepsy versus focal epilepsy.


The AG/DAG ratio was significantly higher in epilepsy patients compared to controls, and it did not differ between different epilepsy types, or between refractory and nonrefractory epilepsy [84]. In females with Rett syndrome, the AG/total ghrelin ratio was significantly increased compared to epilepsy patients not diagnosed with Rett syndrome [87]. An assessment of the ratios of AG and DAG or total ghrelin may explain the difficult-to-reconcile observations in the ghrelin levels, and may appear as a good alternative read-out for these studies. However, measuring AG from plasma is technically challenging, and the best practices for handling AG plasma samples are, up to now, not entirely resolved [88]. Thus, the sensitivity of AG to sample handling can lead to large observed interstudy variations.

To elucidate the impact of ASD treatment or epilepsy disease progression on ghrelin levels, some studies have assessed the interictal ghrelin plasma levels before and after ASD treatment. After two years of successful valproic acid treatment, only patients that had developed obesity had significantly lower plasma ghrelin levels compared to controls, while this was not the case in patients that had not developed obesity [83]. The serum DAG levels did not differ after three months of ASD treatment, while the AG levels were decreased after three months [80]. Finally, ASD-responsive patients had increased ghrelin levels compared to nonresponders in two studies, but not in another study [82,86,89]. A significant positive correlation has been shown between both the AG and DAG levels and disease duration, which could be indicative of ghrelin resistance, but could also be related to ASD use [84].

One study assessed the alterations in plasma ghrelin immediately after seizures. The AG and DAG levels decreased as soon as five minutes after a generalized seizure, and were restored after 24 h [89]. Moreover, in the preclinical pentylenetetrazole (PTZ) model, AG, but not DAG, as well as total ghrelin plasma levels, decreased 30 min after the induction of a seizure (see further) [90]. Overall, most studies show lower ghrelin levels in patients with epilepsy compared to healthy controls, or a decrease in ghrelin levels after a seizure.

#### *3.2. Children*

The latter statement could be extrapolated to children, as total ghrelin levels were significantly lower in prepubertal children with epilepsy compared to healthy controls [91,92]. Another study assessed the AG and DAG plasma levels within six hours after a seizure in children not yet receiving treatment (pretreatment), three months after treatment (posttreatment), and in healthy controls. The AG levels were significantly lower in the pretreatment group compared to the post-treatment group and the controls [93]. DAG levels were significantly higher in the post-treatment group compared to the pretreatment group in urine and saliva, but not in serum [93] (Table 2). Within the epilepsy group, lean children on valproic acid had significantly higher total ghrelin plasma levels compared to children on carbamazepine [91], but not compared to children receiving topiramate [94].

**Table 2.** Overview of interictal ghrelin levels in children. AG: acylated ghrelin; ASD: antiseizure drug; CBZ: carbamazepine; TPM: topiramate; DAG: desacyl ghrelin; Ref: reference; VPA: valproic acid. \* the different ghrelin levels in epilepsy patients versus controls. \*\* the difference in condition 2 versus condition 1 within epilepsy patients. Age denotes either the mean age of the patient groups rounded to the nearest integer, or the age range.


The majority of studies that assessed ghrelin levels related to disease progression did not detect significant differences between ghrelin levels measured over time (Table 3) [95–97]. One study showed that plasma ghrelin was significantly decreased after the initiation of valproic acid treatment in pubertal children, but not in prepubertal children, nor in children on oxcarbazepine. In the latter case, this may be due to the increased weight gain in the children receiving valproic acid [98].

**Table 3.** Overview of interictal ghrelin levels in children after ASD or KD intervention. AG; acylated ghrelin; ASD: antiseizure drug; CBZ: carbamazepine; d: day; DAG: desacyl ghrelin; Int: intervention; KD: ketogenic diet; LEV: levetiracetam; m: month; OXC: oxcarbazepine; PHT: phenytoin; Ref: reference; T: time; TPM: topiramate; VPA: valproic acid; y: year. \* the different ghrelin levels in epilepsy patients over time. ± the concentrations derived from graphs.


The Ketogenic Diet

The ketogenic diet (KD) is an alternative treatment option for refractory epilepsy and has often been proven useful, particularly in children. It remains to be elucidated to what extent the alterations in AG or DAG may mediate some of the effects of the KD [104]. Both AG and DAG levels were shown to be decreased after the initiation of a KD in children with drug-resistant epilepsy [101]. Another study showed that AG plasma levels were decreased as soon as 30 days after the initiation of a KD in children with pharmacoresistant epilepsy [102]. One study did not detect alterations in ghrelin levels after the onset of a KD in drug-resistant epilepsy patients [103] (Table 3).

#### **4. Preclinical Evidence for Ghrelin as a Potential Antiseizure Drug**

Ghrelin, in both its acylated and deacylated form, as well as synthetic ligands, have been studied in rodent seizure and epilepsy models. The majority of these studies focus on the administration of ghrelin or ghrelin-R ligands to modulate seizures or epilepsy, while only a few studies have assessed plasma ghrelin levels in these models.

#### *4.1. Ghrelin in Seizure and Status Epilepticus Rodent Models*

Both systemic and intrahippocampal ghrelin administration were anticonvulsant in the acute rat PTZ model [105–108]. A longer pretreatment of 10 days with ghrelin elicited the same antiseizure effect [107]. One study showed that ghrelin administration in PTZ-treated rats enhanced cognitive capacity in terms of spatial memory [109], which is interesting in light of cognitive impairments as important comorbidities of epilepsy [110].

AG, but not DAG, and total ghrelin plasma levels were decreased 30 min after the induction of a seizure in the PTZ model [90]. This decrease was confirmed in another study, where total ghrelin serum and brain levels were lower in rats after acute PTZ injection, but also after chronic PTZ kindling [111]. Finally, the brain tissue and plasma total ghrelin levels were decreased in mice that exhibited seizures, which were elicited after 24 h of fasting, followed by scopolamine administration [112].

In a study conducted on a rat penicillin model performed under anesthesia, only 1 μg, but not 2 μg of ghrelin administered 30 min after penicillin significantly lowered the spike frequency. These data imply that ghrelin might not follow a linear dose–response curve [113]. There is one study that recently demonstrated ghrelin administration to be proconvulsive in a WAG/Rij rat model presenting with absence seizures, as ghrelin increased the number of spike–wave discharges and the total seizure duration one hour after administration [114].

Ghrelin has been assessed in various status epilepticus (SE) models. However, given the short duration of these experiments, they only reflect the effects of ghrelin on the phenomenon of SE, and not on the subsequent chronic recurrent seizures. Ghrelin was not anticonvulsant in SE models in rats [31,115], except for one study [116], while ghrelin exerted anticonvulsant effects in a pilocarpine tail infusion mouse model and an intrahippocampal kainic acid (IHKA) mouse model [29,116]. One explanation for these observations could be the short timing of ghrelin administration prior to the stimulus (only 10 min) in the rat models. Additionally, the doses that were used in these studies varied highly. Interestingly, it appears that the choice of species may be involved as well, as ghrelin (both at a dose of 0.08 mg/kg and 1.8 mg/kg) was anticonvulsant in mice, but not in the pilocarpine or KA rat model at a dose of 1.5 mg/kg. On the other hand, the effects exerted by ghrelin may not be strong enough to interfere with the development of SE (Table 4).


**Table 4.** Overview of effects of ghrelin in experimental epilepsy models. i.c.v.: intracerebroventricular; i.h.: intrahippocampal; IHKA: intrahippocampal kainic acid; i.p.: intraperitoneal; KA: kainic acid; min: minute; pen: penicillin; pilo: pilocarpine; PTZ: pentylenetetrazole; Ref: reference.

Ghrelin's deacylated form, DAG, was anticonvulsant in the IHKA rat model, the intracerebroventricular pilocarpine rat model, and the pilocarpine tail infusion mouse model [37,115] (Table 5). While this was not the case with ghrelin, the administration of DAG only 10 min prior to a pilocarpine or IHKA stimulus was anticonvulsant [115]. A possible explanation may rely on the fact that DAG has a faster transport rate across the BBB in mice compared to ghrelin [71].

**Table 5.** Overview of effects of desacyl ghrelin in experimental epilepsy models. i.c.v.: intracerebroventricular; i.p.: intraperitoneal; KA: kainic acid; min: minute; pilo: pilocarpine; Ref: reference.


Another possible explanation could be the presence of GOAT in the hippocampus, which may locally acylate extracellular DAG [44]. This leads to the compelling hypothesis

that DAG may exert anticonvulsant effects via its superior brain availability compared to ghrelin, in combination with local acylation, to render AG in the hippocampus and exert anticonvulsant effects via the ghrelin-R. This mechanism would drastically improve ghrelin availability at such difficult-to-reach brain areas. Additionally, this hypothesis may fit into the notion that DAG was shown to require ghrelin-R expression to exert anticonvulsant effects [37].

All preclinical studies involving the effect of ghrelin on seizures or SE have been conducted in male rodents. There is one study that used female rats to investigate whether ghrelin administration differentially affects the incidence of seizures at various time points of the estrous cycle. The authors found that ghrelin was anticonvulsant during all phases of the estrous cycle; however, the effects were more outspoken during the luteal phase compared to the follicular phase [108].

#### *4.2. Ghrelin Receptor Agonists*

A large number of shorter ligands with binding affinity at the ghrelin-R have been synthetized, of which the ghrelin-R agonists, macimorelin, capromorelin, and hexarelin have been tested in animal models of seizures or epilepsy (Table 6).

The pseudotripeptide, macimorelin (H-Aib-(d)-Trp-(d)-gTrp-formyl, also known as "JMV-1843") was first synthesized in 2003 [120], and it is currently on the market for the diagnosis of GH deficiency [121,122]. It is a full agonist of the ghrelin-R and has a longer plasma half-life compared to the endogenous agonist [122,123]. Our group showed that macimorelin was anticonvulsant in both the acute 6-Hz mouse model and fully 6-Hz-kindled mice through the ghrelin-R [124], and in a dopamine 1 receptor (D1R)-mediated mouse kindling model [125]. Macimorelin did not exert anticonvulsant effects in the SE pilocarpine rat model [31,115], but was anticonvulsant in the IHKA mouse model [126]. These studies differed in the dose, the timing of the administration, and the species used, which may explain these conflicting findings.

As ghrelin or macimorelin were shown to exert neuroprotective [28–31] and antiinflammatory effects [31,121,122] in seizure models (see further), our group recently studied whether macimorelin was able to interfere with epileptogenesis. The prevention or attenuation of the development of epilepsy could drastically reduce morbidity and the socioeconomic costs associated with refractory epilepsy. However, we found that macimorelin was anticonvulsive, but not antiepileptogenic, in the IHKA mouse model [126].

Capromorelin is a ghrelin-R full agonist with a high affinity for its receptor [127], and it is currently FDA-approved for veterinary use for increasing food intake [128,129]. Capromorelin was intrahippocampally infused two hours prior to intrahippocampal pilocarpine infusion in rats and decreased the total seizure severity score [116].

The hexapeptide, hexarelin, was developed as a GH secretagogue prior to the discovery of ghrelin [130]. Its potential anticonvulsant effects were assessed in both the pilocarpine rat model and the IHKA rat model. While a low dose (0.33 mg/kg) was anticonvulsant in the pilocarpine rat model, the same administration regimen was not anticonvulsant in the IHKA rat model [115]. This once more underscores the variation between the models and the species used in the discovery of novel potential ASDs, and advocates for the use of multiple seizure or epilepsy models in the discovery of potential new ASDs.

#### *4.3. Administration of Other Ghrelin Receptor Ligands*

Neutral antagonists prevent the activation of a receptor by blocking the agonist binding to the receptor, but do not affect its basal constitutive activity. Three ghrelin-R antagonists have been investigated in a variety of epilepsy models, of which the neutral antagonist, JMV-2959, was without effects in the pilocarpine rat model, in acute 6-Hz- or fully kindled mice, and in the D1R-mediated kindling model [115,124,125]. The hexapeptide, EP-80317 (Haic-D-Mrp-D-Lys-Trp-D-Phe-Lys-NH2), was anticonvulsive in the pilocarpine SE model and the 6-Hz-kindled mouse model [115,131,132] (Table 6). Interestingly, resistance to the initial anticonvulsant effects of EP-80317 treatment were observed with seizure progression

in the 6-Hz-kindling model [131,132]. Its anticonvulsant effects were shown to be dependent on the peroxisome-proliferator-activated receptor, S-gamma (PPAR-γ), presumably via the cluster of differentiation (CD)36 receptor [131,132]. By contrast, one recent study showed that the ghrelin-R antagonist, D-Lys-3-GHRP-6, induced spontaneous seizures in an amygdalakindled rat model, [133].

Intrahippocampal infusion of the inverse agonists, A778193 and [D-Arg1, D-Phe5, D-Trp7,9, Leu11]-substance P, were anticonvulsant in the intrahippocampal pilocarpine infusion rat model. Inverse agonists are typified by their ability to block the intracellular signaling of a receptor, including basal constitutive signaling, which resembles an absence of the receptor. In line with this notion, ghrelin-R knock-out (KO) mice were shown to be protected from seizures [116,124], which suggests that the absence of ghrelin-R signaling is anticonvulsant. In agreement with this, the biased agonist, YIL671, a Gα<sup>q</sup> and Gα<sup>12</sup> selective biased ligand of the ghrelin-R that is not able to recruit β-arrestin, increased the seizure burden in the D1R-mediated kindling model [125].

**Table 6.** Overview of anticonvulsant effects of ghrelin-R ligands in experimental epilepsy models. D1R: Dopamine 1 receptor; i.v.: intravenous; i.h.: intrahippocampal; IHKA: intrahippocampal kainic acid; i.p.: intraperitoneal; KA: kainic acid; min: minute; pilo: pilocarpine; Ref: reference; SP: [D-Arg1,D-Phe5,D-Trp7,9,Leu11]substance P.



**Table 6.** *Cont.*

#### **5. Molecular Mechanisms-of-Action**

*5.1. Mechanisms of Ghrelin's Anticonvulsant Action*

Ghrelin-R expression is dynamic and may be influenced by the presence of a disease state, or may depend on exposure to ghrelin [54,55,134], which are both relevant in the context of ghrelin administration in seizure and epilepsy models. Neither ghrelin nor pilocarpine altered hippocampal ghrelin-R mRNA expression in the pilocarpine rat model [28], while another group showed a decrease in the hippocampal ghrelin-R mRNA expression in pilocarpine-treated rats, which was restored upon ghrelin administration [30].

Hippocampal Akt signaling was decreased in a pilocarpine rat model, which could be restored by ghrelin administration [28,30]. Akt is a downstream target of the ghrelin-R, which can be activated both by Gα<sup>q</sup> signaling and β-arrestin recruitment. The ghrelin-R antagonist, EP-80317, restored the increased hippocampal phosphorylation levels of the other canonical downstream target ERK in the 6-Hz mouse model [131].

Up until now, the exact signaling pathways responsible for anticonvulsant effects downstream of ghrelin-R have remained elusive, but a few possibilities exist, on the basis of previous findings. Not only ghrelin-R agonists, but also ghrelin-R inverse agonists exerted anticonvulsant effects, and ghrelin-R KO mice were protected from seizures [116,124]. The truncated ghrelin variant, ghrelin (1–5) amide, shows similar EC50 values compared to ghrelin with regard to the ghrelin-R signaling pathways, but is unable to internalize the ghrelin receptor, and was not able to exert anticonvulsant effects [116]. Because of these intuitively irreconcilable observations, a novel concept emerged, hypothesizing that the absence of the ghrelin-R on the cell surface was responsible for exerting ghrelin's anticonvulsant effect [116]. We showed that a Gα<sup>q</sup> and Gα<sup>12</sup> selective biased ligand of the ghrelin-R, YIL781, increased seizure severity in a kindling model [125]. Given these observations, β-arrestin recruitment remains the most probable pathway involved in ghrelin-R-mediated anticonvulsive effects and it requires further investigation.

However, we cannot completely exclude that G-protein-dependent signaling may be required for ghrelin-R-mediated anticonvulsant effects, as Akt and ERK activation have been described [30,32,135]. However, this would not fit with the notions of ghrelin-R KO mice being protected from seizures, and inverse agonists exerting anticonvulsant effects; in these cases, there is no G-protein-dependent signaling downstream of the ghrelin-R. Finally, one could hypothesize that the possibility exists that the signaling pathways downstream of β-arrestin may be responsible for ghrelin's anticonvulsant effects. Indeed, Akt and ERK are mediators that can also be activated via β-arrestin-dependent signaling. Nonetheless, and also here, the data obtained from the experiments with ghrelin-R KO mice and inverse agonists [116] suggest otherwise, and point towards an absence of signaling, which is imperative for ghrelin's anticonvulsant effects.

#### *5.2. Mechanisms of Neuroprotection*

Ghrelin increased the number of surviving neurons in the CA1 and CA3 hippocampal regions in the pilocarpine rat model [28]. Pilocarpine reduced the apoptotic repressor, B-cell lymphoma 2 (Bcl-2), increased the proapoptotic member Bcl-2-associated X protein (Bax), and increased cleaved caspase-3, crucial in apoptosis. Ghrelin was able to restore these markers, and may thus exert neuroprotection through antiapoptotic effects [28].

This latter finding was confirmed in the i.p. KA mouse model, in which ghrelin decreased cleaved caspase-3 immunoreactivity in pyramidal CA1 and CA3 neurons, and restored neuronal loss in CA1 and CA3 [29]. Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL)-positive cells were abundantly present in vehicle-treated KA mice, but no TUNEL-positive cells could be observed in ghrelin-treated KA mice. All of the above described effects were dependent on ghrelin-R, as they were reversed by the concurrent administration of a ghrelin-R antagonist [29]. A study conducted by Zhang and colleagues confirmed the necessity of ghrelin-R availability in order for ghrelin to exert its neuroprotective effects [30]. Ghrelin significantly rescued neuronal cell loss in CA3, and inhibited cleaved caspase-3 activation, mediated via the phosphorylation of Akt [30].

A two-week-long administration of the ghrelin-R agonist, macimorelin, in the IHKA mouse model, exerted anticonvulsant effects on spontaneous recurrent seizures, but did not increase neuronal survival in the CA1, CA3, and DG of the hippocampus. This could be due to the omission of pretreatment, as the onset of the treatment commenced 24 h after SE induction, or due to the additional two-week wash-out in this study [126]. A lower dose of macimorelin, administered prior to SE, was found to increase neuronal survival and decrease apoptosis in the DG in the pilocarpine rat model, but not in CA1 [31]. Additionally, this study was able to demonstrate neuroprotective effects exerted by macimorelin, but not anticonvulsant effects against the pilocarpine-induced SE.

The possibility should be considered that ghrelin exerts neuroprotective effects via the activation of a variety of signaling pathways mediated through the employment of Gproteins, whereas the rapid and subsequent internalization via β-arrestin signaling may be responsible for anticonvulsant effects. Unraveling which ghrelin-R downstream signaling pathway is responsible for a particular effect may be obtained by doing further experiments in genetic models, or by using biased ligands that selectively activate a subset of pathways while leaving others untouched.

#### *5.3. Inflammation*

Inflammation is a major hallmark of epileptogenesis, seizures, and chronic epilepsy. This ranges from infiltration of the inflammatory cells and the release of proinflammatory mediators to widespread gliosis [134–137]. Given the fact that inflammation is known to progress the development of epilepsy [138], one of the presumed mechanisms-of-action of ghrelin may rely on its ability to attenuate inflammation, stemming from both direct central actions, and through peripheral anti-inflammatory effects [139].

Ghrelin significantly reduced the elevated plasma calcitonin gene-related peptide (CGRP), substance P, interleukin (IL)-6, tumor necrosis factor (TNF)-α, and IL-1β in the PTZ rat model [117,118]. Additionally, ghrelin inhibited KA-induced increases in TNF-α, IL-1β and cyclooxygenase-2 (COX2) mRNA levels in CA1 and CA3 in the i.p. KA mouse model, mediated via the ghrelin-R [29]. Ghrelin restored KA-induced increased matrix metalloproteinase 3 levels, which is an important mediator of inflammation and neuronal cell death [29]. Additionally, KA-induced increases in microglia and glial fibrillary acidic protein (GFAP) immunoreactivity in CA1 and CA3 three days after SE were inhibited by ghrelin [29]. This was not detectable after a two-week wash-out following macimorelin administration in the IHKA mouse model, in which macimorelin was administered after KA, and not as a pretreatment [126]. Another study showed that ghrelin administration decreased cortical TNF- α and NF-κB expression in the pilocarpine rat model [140]. However, ghrelin did not alter serum levels of galanin, fibroblast growth factor (FGF-2), IL-6, TNF-α, and IL-1β in the Wag/Rij rat model with nonconvulsive absence seizures [114].

#### *5.4. Oxidative Stress*

Seizures induce oxidative stress, which, in turn, exacerbates seizures (reviewed by [141]). Ghrelin prevented the PTZ-induced decrease in the catalase activity in both the CNS and erythrocytes, and prevented the augmentation in thiobarbituric acid reactive substances levels, a measure for lipid peroxidation [142]. Additionally, ghrelin normalized superoxide dismutase levels, an enzyme responsible for clearing superoxide anion in the erythrocytes, brain, and liver [142]. These data suggest that ghrelin protects against oxidative stress caused by PTZ. It remains unclear whether the effects of ghrelin on decreasing oxidative stress are caused directly, or because ghrelin is anticonvulsant, and the lower number of seizures leads to decreased oxidative stress. However, in the WAG/Rij rat model with nonconvulsive absence seizures, ghrelin was not anticonvulsant, but it still reduced the malondialdehyde [114].

#### **6. Functional Implications of Diminished Ghrelin-R Signaling in the Context of Excitability**

While several molecular mechanisms-of-action have been described, it remains unknown how these contribute to the anticonvulsant effects of ghrelin, and how they lead to an overall decrease in the brain excitability. The heteromerization of the ghrelin-R with other receptors can lead to the preferential recruitment of other noncanonical signaling pathways [143]. Ghrelin, as well as other signaling molecules, may exploit this phenomenon for inducing ghrelin-R-mediated anticonvulsant effects.

Ghrelin-R activation results in intracellular Ca2+ increases through the canonical Gα<sup>q</sup> protein [59,60]. One possible mechanism-of-action would be a decrease in intracellular Ca2+ in ghrelin-R-expressing neurons. Elevated levels of intracellular Ca2+ are associated with epileptiform activity and epileptogenesis [144,145]. Therefore, a reduction in intracellular Ca2+ may be an interesting putative mechanism for seizure suppression in the absence of ghrelin-R signaling. Various studies have shown the differential effects of ghrelin on neuronal excitability and synaptic transmission, which all support that ghrelin acts in a brain-region-specific manner [146–148].

The ghrelin-R is expressed in both excitatory neurons, as well as in inhibitory interneurons in the dorsal CA1. It was recently shown that a selective increased expression of the ghrelin-R in excitatory neurons was detrimental for learning and memory in mice, while an increased expression of the ghrelin-R in interneurons had a beneficial effect [149]. It remains to be uncovered if a dual effect on excitability also exists by, for instance, decreasing Gα<sup>q</sup> signaling in excitatory neurons via β-arrestin, while increasing Gα<sup>q</sup> signaling in inhibitory interneurons. Indeed, while GPCRs may be associated with several signaling pathways, these signaling pathways are not always all operative in the same cell. Thus far, the knowledge concerning the cell-specific expression of signaling pathways downstream of the ghrelin-R is lacking and requires further studies.

#### **7. Conclusions and Future Perspectives**

Ghrelin is increasingly recognized as a potential important player in seizures and epilepsy. Most studies show lower ghrelin levels in patients suffering from epilepsy, or lower ghrelin levels after a seizure. The exact implications of plasma ghrelin level alterations in epilepsy have remained, up until now, unknown, and should be further investigated in light of its treatments as well, including the KD. It is increasingly evident that there may be important differences between AG and DAG. This may advocate for future investigations of both isoforms of ghrelin in epilepsy, or for studying further whether the contributions of GOAT expression and the local reacylation of DAG are relevant for seizure control.

With only a few exceptions, ghrelin and synthetic agonists of the ghrelin-R are anticonvulsant in seizure, epilepsy, and SE models. The notion that both agonists and inverse agonists were anticonvulsant stirred up the discussion concerning the signaling pathways responsible for ghrelin-R-mediated anticonvulsant effects. The hypothesis that β-arrestin recruitment is involved should be more thoroughly investigated to confirm the relevance of this pathway. Overall, the complexity of ghrelin-R signaling, and the extensive list of other factors possibly influencing it, highlight the need for further investigations into the mechanism behind ghrelin-induced anticonvulsant effects.

**Author Contributions:** Conceptualization, A.B., D.D.B., R.K. and I.S.; Investigation, A.B.; Resources, D.D.B., R.K. and I.S.; Writing—Original Draft Preparation, A.B.; Writing—Review & Editing, A.B., D.D.B., R.K. and I.S.; Visualization, A.B.; Supervision, D.D.B., R.K. and I.S.; Project Administration, A.B.; Funding Acquisition, A.B., D.D.B., R.K. and I.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the scientific Willy Gepts fund of the UZ Brussel, the Queen Elizabeth Medical Foundation (ING prize), and the strategic research program of the Vrije Universiteit Brussel (SRP49). A Buckinx is a research fellow of the Fund for Scientific Research, Flanders (SB-FWO grant no. 1S84218N).

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

#### **References**


#### MDPI

St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Biomedicines* Editorial Office E-mail: biomedicines@mdpi.com www.mdpi.com/journal/biomedicines

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel: +41 61 683 77 34

www.mdpi.com ISBN 978-3-0365-6437-1