Authors: Masaya Togo (1Laboratory of Behavioral and Cognitive Neuroscience, Stanford School of Medicine, Palo Alto, CA 94305, USA), Dian Lyu (1Laboratory of Behavioral and Cognitive Neuroscience, Stanford School of Medicine, Palo Alto, CA 94305, USA), Weichen Huang (1Laboratory of Behavioral and Cognitive Neuroscience, Stanford School of Medicine, Palo Alto, CA 94305, USA), Sofia Pantis (1Laboratory of Behavioral and Cognitive Neuroscience, Stanford School of Medicine, Palo Alto, CA 94305, USA), Robert Fisher (2Department of Neurology and Neurological Sciences, Stanford School of Medicine, Palo Alto, CA 94305, USA), Riki Matsumoto (3Department of Neurology, Kyoto University, Graduate School of Medicine, Sakyo-ku, Kyoto 6068507, Japan), Vivek Buch (4Department of Neurosurgery, Stanford School of Medicine, Palo Alto, CA 94305, USA), Josef Parvizi (1Laboratory of Behavioral and Cognitive Neuroscience, Stanford School of Medicine, Palo Alto, CA 94305, USA; 2Department of Neurology and Neurological Sciences, Stanford School of Medicine, Palo Alto, CA 94305, USA; 4Department of Neurosurgery, Stanford School of Medicine, Palo Alto, CA 94305, USA)
Categories: Article, epilepsy, thalamo-cortical connectivity, evoked potentials, neuromodulation, intrathalamic, interthalamic
Source: Brain : a journal of neurology
Authors: Masaya Togo, Dian Lyu, Weichen Huang, Sofia Pantis, Robert Fisher, Riki Matsumoto, Vivek Buch, Josef Parvizi
The Papez circuit traditionally highlights the anterior nuclei of the thalamus (ANT) as the main relay of hippocampal (HPC) output to the cortex, a view that has shaped neuromodulation strategies in temporal lobe epilepsy (TLE). However, recent studies suggest that the medial subregion of the pulvinar (mPLV)—a thalamic nucleus that has undergone significant evolutionary expansion throughout the mammalian brain evolution—also forms functional connections with medial temporal lobe (MTL) structures, including the HPC. To date, however, there is a lack of causal evidence directly comparing the connectivity between the HPC and the two thalamic nuclei (mPLV and ANT) and between the two thalamic structures within the same brains.
In this study, we investigated 41 patients with medial (mTLE, n = 22) and non-medial temporal lobe epilepsy (non-mTLE, n = 19) implanted with simultaneous depth electrodes in the HPC, ANT and mPLV. Repeated single-pulse electrical stimulations were applied to compare the causal electrophysiological connectivity of these regions within the same individuals.
Our intra-subject analysis revealed that anterior HPC stimulation evoked strong responses in both ANT and mPLV, with mPLV responses occurring significantly later than those in the ANT [linear mixed-effect model (LMM), 10.19 ms, 95% confidence interval (CI) (1.78, 18.59), false discovery rate (FDR)-corrected P = 0.040]. In contrast, stimulation of the posterior HPC resulted in stronger (and a trend for earlier) responses in mPLV compared with ANT [LMM: 0.117, 95% CI (0.033, 0.201) (FDR-corrected P = 0.012)].
This finding suggests an anterio-posterior gradient of HPC connectivity. Furthermore, we found robust bilateral and bidirectional connectivity between ANT and mPLV. Stimulation of either elicited responses in the other, including the contralateral thalamus. This represents clear evidence for both intrathalamic and interthalamic connectivity within the human brain. Our findings offer new insights about the connectivity of human HPC with the thalamus and strong intrathalamic exchange of electrophysiological activity within the human brain.
Deep brain stimulation (DBS) of the thalamus is an emerging new strategy for treating medically refractory epilepsy.^1^ While DBS of the anterior nuclei of the thalamus (ANT) has proven to be effective in well-controlled randomized clinical studies,^2^ the real-life efficacy of this approach, as shown in observational registries of prospective and retrospective data, has been less impressive.^3–5^ To date, it remains unclear if targeting other thalamic nuclei could be more suitable for patients with epilepsy.
Targeting the ANT, especially in patients with temporal lobe epilepsy (TLE), is a strategy based on a long-held neuroanatomical belief that the main outlet of the hippocampus (HPC), hence the main outlet for propagation of hippocampal seizures, is through the Papez circuit.^6^ Emerging evidence from comparative neuroanatomical data^7–12^ as well as imaging^13,14^ and electrophysiological connectivity studies^15^ suggests that the medial pulvinar (mPLV) ought to be considered a key thalamic target for the hippocampal output apart from the Papez circuit.
In line with extant anatomical and physiological evidence, several observational studies in epilepsy patients implanted with stereotactic EEG (sEEG) electrodes have suggested the engagement of the thalamus during hippocampal seizures. In first-of-its-kind recordings directly from the human thalamus, Bartolomei’s team in France obtained single-site recordings from the thalamus and documented its involvement in the majority of patients.^16^ These findings were extended to ANT, mPLV and other thalamic subregions.^17–24^ While these studies unanimously documented the involvement of the thalamus during early phases of seizure propagation, they were mostly focused on the engagement of single thalamic sites and did not compare the time or predominance of seizure propagation through specific thalamic sites at the individual patient level. Inspired by the methodology used in prior studies (i.e. extending clinical electrodes from cortical to subcortical regions to reach the thalamus),^18–20,25–27^ we addressed the gap of knowledge by collecting simultaneous and bilateral multi-site recordings within the thalamus in each patient. We reported that the engagement of the mPLV was either earlier or simultaneous with the engagement of the ANT in the majority of seizures in patients with TLE.^28^ In a follow-up study, we reported that in most patients, both ANT and PLV were involved during seizures, but in 82% of the patients, seizure spread was noted first in the PLV. In patients with confirmed hippocampal/amygdala onset seizures, 62% had initial involvement of the PLV and 100% had subsequent involvement. Only 31% showed initial propagation to ANT.^29^ In line with these findings, recent DBS of the PLV in a series of patients has shown promising results for controlling seizures.^30^
The existing evidence, as reviewed above, supports the hypothesis that the HPC has stronger causal connections with the mPLV than with the ANT in the human brain. However, this hypothesis has not been directly tested through a systematic comparative study at the individual patient level. Our recent study explored causal thalamo-cortical connectivity using cerebro-cerebral evoked potentials (CCEP),^15^ which was originally referred to as ‘cortico-cortico evoked potentials’ but will be henceforth referred to as ‘cerebro-cerebral evoked potentials’ as it involves non-cortical thalamic seeds and targets,^31^ and is less prone to corticocentric myopic bias.^32^ This work introduced a novel data-driven approach for classifying CCEP and demonstrated widespread evoked responses in both ipsilateral and contralateral cortices with an antero-posterior gradient (Fig. 1). However, a comparative connectivity analysis between the HPC and the two thalamic structures was not conducted.
Previous CCEP studies investigating hippocampothalamic connectivity have been limited,^33,34^ with a small number of patients and an absence of direct comparison between the ANT and mPLV at the individual brain level. The goal of the present study was to address this knowledge gap and perform a head-to-head comparison of causal connectivity between the HPC, ANT and mPLV using CCEP.
In total, 41 patients [17 (41.4%) female; mean age ± standard deviation (SD): 37.6 ± 12.0 years] were recruited. All patients were diagnosed with focal epilepsy, and the sEEG electrodes were implanted at Stanford Health Care from 2021 to 2024 (Supplementary Table 1). All patients were implanted in the anterior and/or posterior part of the thalamus to identify seizure onset and possible candidates for neuromodulation therapy. Each participant had 188 ± 46 (mean ± SD) sEEG contacts implanted. The total number of electrode contacts included in this study is 7728. Patients were classified into two an mTLE group (n = 22) and a non-mTLE group (n = 19) based on the location of confirmed seizure onset zones (Supplementary Table 1).
The placement of electrodes was determined based on clinical requirements. The strategy of thalamic implantation was discussed and approved by all epileptologists, neurosurgeons, neuroradiologists and neuropsychologists based on pre-surgical evaluation during weekly surgery meetings. All participants provided written informed consent, and the Institutional Review Board approved the study protocol for human experimentation (protocol 11354). Research procedures and consents were approved by the Stanford IRB as per routine clinical protocols.
We used a reduced diameter obturating stylet and reduced diameter electrodes with 0.86 mm diameter (AdTech Medical) for the thalamic recording, which is in line with our previous works.^15^ Interelectrode distances were 5 mm. In Patient 41, the thalamic recording electrodes were made in another company (PMT corporation). Their diameters were 0.8 mm and interelectrode distances were from 3.5 to 4.43 mm. Trajectory planning has been discussed in our previous study.^35^ The anterior and posterior parts of the thalamus were reached by extending the electrodes implanted for sampling targeted cortical regions of interest. The anterior trajectory extended from the frontal or anterior temporal to anterior insular to anterior thalamus, while the posterior trajectory extended from the posterior temporal operculum, temporoparietal junction or supramarginal gyrus to the posterior insular to the posterior thalamus. No additional electrodes were needed for thalamic recordings. Trajectories avoided middle cerebral vessels and minimized the distance travelled through the Sylvian fissure.
A thin-cut head CT, obtained after electrode implantation, was co-registered to pre-operative MRI data to verify the trajectory. Surface reconstruction was generated from the pre-operative T1-weighted MRI scan by using the recon-all command of Freesurfer v.7.3.2.^36^ The post-implant CT scan was co-registered with the pre-implant MRI using the flirt function from the Oxford Centre for Functional MRI of the Brain Software Library.^37,38^ We manually identified each electrode on the T1-registered CT image using BioImageSuite.^39^ The anatomical location of the electrodes in the native space were manually labelled by a trained neuroanatomist based on the individual brain’s morphology and landmarks. Contacts entirely or partially located outside of the grey matter were excluded from the analysis. For group-level visualization, the electrode coordinates in the native space were converted to the FSaverage or MNI305 space by using the iElVis toolbox.^40^
In the present study, we focused on the ANT and the mPLV. We used the Thalamus Optimized Multi Atlas Segmentation (THOMAS) automated pipeline to segment the thalamus on an individual basis using each participant’s own T1 image (https://github.com/thalamicseg/hipsthomasdocker).^41^ Since the parcellation of mPLV was not defined in the THOMAS atlas, we used the Freesurfer atlas^42^ to obtain the mPLV parcellation.^30^ For identifying the thalamic location of the electrodes across electrodes, 5 mm spheres were created around the inner contact centres of mass (contact neighbourhood), and a binary mask with the spheres was taken as the affected area by electrical stimulation. The fraction of voxels in the binary mask that overlapped with the segmented thalamic nuclei was calculated. We set the value to 5 mm based on our previous study,^43^ with modifications due to differences in interelectrode distances compared with the previous dataset.
Regarding CCEP processing, we followed our previous work, which describes the methodology in detail.^15^ Low-frequency (0.5 Hz) electrical stimulations were delivered in a bipolar manner between electrode pairs. Each stimulation had a pulse duration of 0.2 ms, with a current of 6 mA (4 mA near seizure onset zones). Each CCEP trial consisted of 42 ± 2 stimulations. Data exclusion criteria (i) bipolar channels in white matter; (ii) recording sites within 5 mm of the stimulation site; (iii) channels with high artefact contamination (≥7 SD of baseline within <10 ms); and (iv) bad channels identified through predefined criteria, including unstable impedances or excessive noise. To ensure reliable CCEP measurements, we assessed stimulation-related artefacts and removed instances where artefacts extended beyond the expected contamination window. Stimulation intensities were adjusted for specific electrode locations to minimize the risk of seizure induction.
Intracranial EEG data were recorded using the Nihon Kohden system (1000 Hz sampling rate) and preprocessed with an in-house pipeline. Preprocessing included notch filtering (60, 120, 180 Hz), exclusion of pathological/noisy channels and bipolar re-referencing. Pathological channels were identified by high-frequency oscillations (HFOs),^44^ while noisy channels were detected by extreme amplitude [≥5 standard deviations (SD) across all channels] and spike prevalence (≥3 median of the distribution across all channels). Trials with extreme mean amplitude (≥4 SD across all trials) were also excluded.
Wavelet decomposition [Morlet wavelets, log-spaced 1–256 Hz (59 total frequencies), each wavelet having 5-cycle width] was applied, yielding a power timeseries with a sampling rate of 200. Data were epoched, baseline-corrected and averaged across good trials, and then values in the power spectrogram were log-transformed and Z-scored across time and frequency. To measure how CCEPs are consistent across trials, inter-trial phase coherence (ITPC) was computed, square-root transformed and Z-scored across time and frequencies. Instead of using common average referencing, we used bipolar referencing as it preserves localized signals. As a final step in our quality control, we manually inspected the raw data and ensured that neurophysiological responses were not distorted.
Details of our method are described in Lyu et al.^15^ In brief, our novel method was developed to overcome a significant problem in the field of CCEP research, namely the dependence on arbitrary thresholds and amplitude-based criteria to differentiate significant evoked responses from noise. With a few exceptions,^45,46^ previous studies have been largely reliant on simple univariate measures detecting large peaks or the time-to-peak in the evoked signals, and limit responses to fixed windows of interest. These traditional methods are not able to capture the complex dynamics of physiological responses generated by electrical stimulations, and they can vary significantly depending on the chosen cut-off Z-scores for determining an effect, current intensity, and the distance between stimulating and recording electrode contacts.^47^ These problems may become more significant since metrics of ‘significant’ evoked response have not been validated on subcortical data.
To bypass the existing methodological hurdles, we identified the evoked electrophysiological effects of electrical stimulation by combining a non-linear manifold learning algorithm supervised with human judgement, using time–frequency spectral features of the evoked potentials as the input data. We initially labelled data tentatively based on two connectivity (i) trial-averaged power (i.e. response strength); and (ii) ITPC (i.e. response consistency). A semi-supervised machine learning approach then refined these labels, clustering each subject’s data into ‘activated’ (i.e. there is a significant evoked response) and ‘non-activated’ (i.e. there is not). The true evoked responses were distinguished from spontaneous activity and stimulation artefact by identification of large but smooth changes in power and ITPC from baseline, with the changes being consistent over trials. Final classifications were verified by visual inspection of original and reconstructed data. Then using the data from the ‘activated’ cluster, we explored their electrophysiological properties. A dimensionality reduction algorithm revealed distinct data structures, represented by spectral features that were significantly different among each other. These spectral features are shown as non-overlapping frequency and time windows on the group-level power and ITPC spectrogram, averaged among the individual evoked spectrogram of each cluster. As the features were not overlapping in the spectrograms, we could further extract when and where they show up in the stimulated spectrograms of a specific site by using a sliding window approach. The resulting time series of correlation coefficient (r) had a sampling rate of 200 per second, the same as the spectral dataset. A sliding-window cross-correlation with the identified neural feature generated temporal similarity curves for each stimulated-recoded instance, each corresponding to the neural feature’s temporal evolvement in a specific pair of stimulated/recorded sites. With the temporal similarity curve, we detected the peak of the similarity curve (peak_maxCor) and the delay to the peak (time_maxCor) using MATLAB ‘findpeak’ function.
To quantify the connectivity between the thalamus and hippocampus, or among thalamic nuclei, we used peak_maxCor (r) in F1 features. We compared the peak_maxCor values and latencies (time_maxCor) among the connectivity from the thalamus (ANT and mPLV) to the HPC (anterior or posterior HPC), as well as from the HPC to the thalamus. We divided the HPC into anterior and posterior regions based on the Montreal Neurological Institute (MNI) y-coordinate of −22, as per a previous study.^48^ We calculated the response occurrence rate for inflow and outflow by the number of significant responses divided by the total number of stimulation-recording electrode pairs for each patient. These values were averaged across all patients for F1 features. To examine the correlation between the connectivity strength and HPC antero-posterior axis, we computed Pearson’s correlation coefficient between r and the MNI y-coordinates of HPC electrodes. Correlations were assessed separately for both ANT and mPLV.
For inter- and intrathalamic connectivity, we examined the connections from the ANT to the mPLV and the mPLV to the ANT. The thalamic nuclei were classified into the left ANT (LANT), right ANT (RANT), left mPLV (LmPLV) and right mPLV (RmPLV). We calculated the response occurrence rate for each pathway (i.e. LANT-RmPLV, RANT-LmPLV, LmPLV-RANT, RmPLV-LANT and LANT-LmPLV, RANT-RmPLV, LmPLV-LANT, RmPLV-RANT) by dividing the number of significant responses by the total number of stimulation-recording electrode pairs for each patient. These values were averaged across all patients for F1 features.
To compare the connectivity strength (r) and latency between the ANT and mPLV, we applied a linear mixed model (LMM) for model fitting, using the thalamic nuclei (differences between ANT and mPLV) as a fixed effect and subject and recording electrodes as random factors. As a sensitivity analysis to assess whether the observed differences varied by epilepsy subtypes (mTLE versus non-mTLE), we performed LMM including epilepsy subtype in addition to the thalamic nuclei (ANT versus mPLV) and the interaction term was investigated.
We use the generalized linear mixed model (GLMM) for response occurrence rate and inter- and intrathalamic connectivity. We set the pathway (inflow or outflow) and thalamic nuclei as fixed effects and subject and recording electrodes as random factors in analysing the response occurrence rate between the HPC and the thalamus, while we used thalamic nuclei as a fixed effect and subject as random factors in the analysis of inter-intrathalamic connectivity.
Based on a hypothesis testing framework, the model comparison approach was used for significance inference. Specifically, a null model was compared with an alternative model that included the hypothesized effect added on top of the null model. The model with higher evidence was selected using a likelihood-ratio test. Model fittings and selections were performed using the lme4^49^ and lmertest^50^ toolbox and the anova function in R. The difference in F1 inflow (HPC to thalamus), F1 outflow (thalamus to HPC) were investigated. We adjusted the multiple comparisons using a false discovery rate (FDR).
Before reporting the anatomical location of the electrodes, two key issues need to be emphasized. First, we acknowledge that the location of the electrodes often shift when we transfer the anatomical information from the native to standard atlas space. Second, our study involved electricity and electrical fields (in terms of either recording electrophysiological evoked responses or injecting electrical pulses to stimulate a region). Hence, we are mindful that the reported anatomical locations and the actual physiological spaces affected are most likely not entirely concordant.
In all patients, the ANT and mPLV electrodes were located in the clinically intended and neuro-surgically planned locations. However, when we transferred from native space to standard atlas space, based on the THOMAS atlas, the ANT electrodes were located in the anteroventral (AV) nucleus in 33.9% of cases (20/59), the ventroanterior (VA) nucleus in 47.4% (28/59) and the ventrolateral (VL) nucleus in 8.4% (5/59). We emphasize that these locations may not be the same as the actual targets of stimulations or recordings due to the two reasons stated above. Also we emphasize that the size of electrodes implanted in patients with epilepsy (e.g. for chronic DBS) are large relative to the small size of ANT, and hence, the DBS in target populations most likely affects a larger segment of the anterior thalamus. For parcellating the mPLV, we used the Freesurfer atlas,^42^ as the THOMAS atlas does not include the mPLV. According to this atlas, 88.1% (52/59) of the innermost posterior thalamic electrodes were localized to the mPLV.
The location of the implanted electrodes across all 41 subjects is illustrated in Fig. 1. In total, 257 electrodes were implanted in the HPC (174 in the anterior and 83 in the posterior half of the structure), 118 in the ANT and 118 in the mPLV (Fig. 1C and D).
A total of 17 patients underwent anterior HPC stimulation with recordings in both ANT and mPLV. Among them, 13 patients exhibited responses in both the ANT and mPLV, while one patient showed a response only in the ANT and three patients only in the PLV. For posterior HPC stimulation, 17 patients (not the same cohort as the anterior HPC stimulation) were also evaluated. Of these, 12 patients had responses registered in both the ANT and mPLV, while three patients showed responses only in the ANT and two patients only in the mPLV (Fig. 2C). The occurrence rates of significant responses are shown in Fig. 3. The response rates of inflow are significantly larger than those of outflow [GLMM, odds ratio = 0.073, 95% CI (0.072, 0.074), P < 0.001]. There were no significant differences between the ANT and mPLV in the proportion of inflow and outflows.
Using the F1 signal as the signal of choice and ‘r’ as the measure of strength (see the ‘Materials and methods’ section), we conducted within-individual comparison of inflow and outflow connections. We found that the strength of inflow connectivity (i.e. incoming signals) to the ANT and mPLV was different for the anterior versus posterior HPC. Inflow connections from the posterior HPC were significantly stronger to the mPLV than ANT [LMM, estimated mean difference = 0.117, 95% CI (0.033, 0.201), FDR-corrected P = 0.012; Fig. 2A]. Inflow connections from the anterior HPC to the ANT and mPLV were statistically similar. By contrast, outflow connections from the ANT were stronger than those from the mPLV to the anterior HPC [LMM, estimated mean difference = −0.219, 95% CI (−0.328, −0.110), FDR-corrected P = 0.0022; Fig. 2A] while the outflow connections of both to the posterior HPC were statistically similar. The connectivity strength of the mPLV were significantly correlated with the antero-posterior axis of HPC electrodes (MNI y-axis coordinate), both inflow and outflow (Pearson’s r = −0.495, P < 0.001, Pearson’s r = −0.338, P < 0.05, respectively). There is not such significant correlation between the connectivity strength of the ANT and antero-posterior axis of HPC electrodes.
Within-individual latency analysis showed that the anterior HPC stimulations evoked earlier F1 responses in the ANT than mPLV [LMM, 10.19 ms, 95% CI (1.78, 18.59), FDR-corrected P = 0.040; Fig. 2B]. There were no significant differences in the latency of F1 inflow connections from the posterior HPC to ANT versus mPLV [LMM, −7.34 ms, 95% CI (−16.29, 1.61), FDR-corrected P = 0.103; Fig. 2B], and there was no difference in the latency of outflow connections from the ANT versus mPLV to different segments of the HPC.
We separated the cohort into mTLE and non-mTLE groups based on the location of the seizure onset zones identified in each patient (Supplementary Table 1). We performed the same analysis regarding the r-values across the two. Because the number of observations was unequal across pairs of interest (stimulation and response sites) between the two groups, we used LMM analysis including thalamic nuclei (ANT and mPLV) and epilepsy subtypes (mTLE and non-mTLE) in a sensitivity analysis to assess whether the observed differences varied by epilepsy subtypes. None of the interaction terms (inflow/outflow between the thalamus and anterior/posterior HPC) reached statistical significance (P > 0.1), suggesting that the effect is not significantly different between mTLE and non-mTLE groups. We visually plotted the comparisons of interest across the two groups and found similar trends in both groups supporting the main findings of the study (Supplementary Fig. 1).
Figure 4A illustrates an example waveform of intrathalamic connectivity in Patient 41. Stimulation of the ANT elicited multiphasic responses lasting approximately 600 ms in the mPLV (Fig. 4A, lighter line). Similarly, stimulation of the mPLV induced evoked responses in the ANT (Fig. 4, darker line). A total of 129 responses (ANT-mPLV pairs) were recorded as inter/intrathalamic connectivity. The occurrence rates of F1 responses were 52.3% to 76.2% for interthalamic (L ANT and RmPLV, RANT and LmPLV, LmPLV and RANT, RmPLV and LANT) connectivity. In comparison, it ranged from 60.9% to 88.5% for intrathalamic (i.e. ipsilateral connections) (Fig. 4C). The intrathalamic occurrence rates of F1 responses were significantly larger than the interthalamic ones [GLMM, odds ratio = 2.62, 95% CI (1.02, 6.77), P = 0.046].
The present study offers the first comprehensive comparison of thalamo-hippocampal connectivity, focusing on the ANT and mPLV in the thalamus. Our findings are based on a novel way of analysing connectivity through CCEP—as we have recently described elsewhere^15^—moving away from arbitrary thresholds of significance and encompassing the signals within the whole spectrum of EEG bands.
As expected, we confirmed that the occurrence rates of thalamic responses evoked by the stimulation of the HPC were significantly more frequent than occurrence rates of HPC responses evoked by the stimulation of either of the thalamic nuclei. Surprisingly, however, our findings offer new insights into distinct connectivity patterns between the human HPC and the two studied thalamic the mPLV receives stronger inputs from the posterior HPC and as such one can hypothesize that it is the thalamic counterpart for the posterior hippocampal processes. Moreover, the abundant inter- and intrathalamic responses observed in our study suggest that the stimulation of a given thalamic site evokes strong signals in other thalamic nuclei even the ones located on the contralateral thalamus.
Our findings are in keeping with anatomical data in rodent^6,51,52^ and non-human primate studies that have documented direct projections from the HPC to ANT^11^ and from the ANT to HPC and subiculum.^12,53^ Additionally, the ANT may connect to the HPC indirectly through the posterior cingulate cortex^45^ or retrosplenial cortex,^54,55^ making these two direct and indirect connections the anatomical pathways for the ANT-HPC effective connectivity observed in our study.
While the ANT-HPC connections are well studied, the hippocampal connectivity with the mPLV has been less understood, especially in primates. In animal studies, PLV has primarily been associated with the occipital and parietal lobes.^7–9^ However, there is scarce evidence for non-fornical connectivity between the HPC and mPLV possibly through the bundle of Arnold^11^ or amygdala.^10^ Imaging studies in humans have also documented functional connectivity between the PLV and HPC but the directionality of this functional relationship has remained unclear.^14^ Our findings are also in agreement with a prior study of a smaller number of patients in which the stimulation of the PLV in two patients evoked responses in the HPC while the stimulation of the HPC in four of five patients evoked responses in the PLV.^33^ Another study also demonstrated connectivity between the PLV and HPC; however, no comparisons were made with other thalamic nuclei.^34^
Our finding of a closer relationship between the posterior HPC and mPLV is also in keeping with the extant literature documenting a role in spatial navigation and visuospatial processing for both of these structures,^14,56,57^ and a strong functional and structural connectivity between posterior medial temporal lobe structures and posterior cortical regions.^58,59^ We recently showed a posterior gradient for cortical areas connected with the mPLV and other PLV subregions compared with those connected with the ANT (Fig. 1B). In patients with TLE, seizures are often thought to originate from the anterior HPC^60^ and surgical resection of the anterior temporal lobe has been effective in controlling seizures.^61^ However, in a recent study, it was shown that targeting the fasciola cinereum of the hippocampal tail was an effective target for controlling seizures.^62^ Lastly, we believe that the posterior HPC involvement in TLE patients, along with the stronger connectivity between the mPLV and the posterior HPC, as documented here, may explain the recent findings that the PLV is involved as early and as prominently as the ANT in seizures originating from the mesial temporal lobes.^28,29^ Based on these findings, it is reasonable to assume that the mPLV could be a useful neuromodulatory target in patients with epilepsy as recently proposed.^30,63^
The causal electrophysiological relationship between the posterior and anterior nuclei of the human thalamus, both ipsilaterally and contralaterally, as documented here, deserves special attention. One candidate structure for contralateral connectivity is massa intermedia, or interthalamic adhesion, which might be a potential pathway for bilateral exchange of information across the two thalami.^64^ However, this structure is not present in all brains.^65^ Another candidate for intrathalamic connectivity is the reticular nucleus of the thalamus, which is known to play a major role in coordinating physiological activity across cortical areas and thalamic nuclei as well as across the thalamic nuclei themselves.^66,67^ While our coarse approach was not suitable to elucidate the precise pathways of the intrathalamic relationship, our findings clearly suggest that the stimulation of ANT in patients with thalamic DBS certainly affects the ipsilateral and contralateral pulvinar structures and thus the effect of thalamic DBS targeting one small nucleus may be more wide-spread than previously anticipated.
We acknowledge that our study had several limitations. First, applying electrical current in the structures between two adjacent electrode contacts (corresponding to 3–5 mm between the two contacts), certainly involves neuronal populations beyond this smaller targeted space. Hence, the electrophysiological connections, as documented here, cannot be taken as precise connections between histochemically defined neuronal structures but rather between approximate regions and fields. Second, our connections are based on the CCEP approach and may not translate to normal physiological relationships between two functional units within the brain. Hence, future studies in primates and human subjects are needed to address these issues to further clarify the implications of our bi directional thalamo-hippocampal CCEP connectivity for seizure propagation and neuromodulation. Despite its limitations, our study provides novel information about the profile of electrophysiological interactions between the ANT and mPLV, and between both of these thalamic structures and the HPC, which deserve future replications.
Supplementary material is available at Brain online.