Integrative roles of human amygdala subdivisions: Insight from direct intracerebral stimulations via stereotactic EEG

Substantial studies of human amygdala function have revealed its importance in processing emotional experience, autonomic regulation, and sensory information; however, the neural substrates and circuitry subserving functions have not been directly mapped at the level of the subnuclei in humans. We provide a useful overview of amygdala functional characterization by using direct electrical stimulation to various amygdala regions in 48 patients with drug‐resistant epilepsy undergoing stereoelectroencephalography recordings. This stimulation extends beyond the anticipated emotional, neurovegetative, olfactory, and somatosensory responses to include visual, auditory, and vestibular sensations, which may be explained by the functional connectivity with cortical and subcortical regions due to evoked amygdala‐cortical potentials. Among the physiological symptom categories for each subnucleus, the most frequently evoked neurovegetative symptoms were distributed in almost every subnucleus. Laterobasal subnuclei are mainly associated with emotional responses, somatosensory responses, and vestibular sensations. Superficial subnuclei are mainly associated with emotional responses and olfactory and visual hallucinations. Our findings contribute to a better understanding of the functional architecture of the human amygdala at the subnuclei level and as a mechanistic basis for the clinical practice of amygdala stimulation in treating patients with neuropsychiatric disorders.


| INTRODUCTION
The amygdala is an almond-shaped structure found in the cerebral hemisphere of all vertebrates. It comprises a dozen nuclei and is implicated in several specific functions, including stimuli perception, emotion, cognition, and environment-adapted behavior (Urban & Rosenkranz, 2020). The amygdala has been implicated in several human and animal diseases, such as psychological depression (Linhartová et al., 2019), schizophrenia (Zheng et al., 2019), dementia (Kark & Kensinger, 2019), chronic pain (Thompson & Neugebauer, 2017), and substance abuse (Munshi et al., 2021). Therefore, it is crucial to obtain information on the physiology and function of the amygdala, which is challenging in humans because its role in healthy individuals and the development of specific diseases is complex and critical.
Animal studies have revealed that the mammalian amygdala contains a constellation of subnuclei (Duvarci & Pare, 2014;Sah et al., 2003), which were composed of at least three subdivisions, including the basolateral complex of the amygdala (BLA), medial amygdala, and central amygdala (CeA) (Ehrlich et al., 2009;McDonald, 2003), with different functions and connections. Internuclei integration and connectivity between subnuclei and the cortical cortex are usually flexible and critical for behavior guidance (Gothard, 2020). The BLA receives sensory information from cortical and thalamic projections, whereas the CeA is the major output of the amygdala (Duvarci & Pare, 2014). Thus, subnuclei-level research is essential to profile amygdala function.
The human amygdala consists of six subregions based on its cytoarchitecture: the laterobasal group (LB), the superficial group (SF), the centromedial group (CM), the intermediate fiber bundles (IF), the ventromedial part (VTM), and the medial fiber bundles (MF) (Amunts et al., 2020). Several attempts have been made to define the functions of the amygdala subnuclei in human mood, autonomic nervous regulation, and internal states (Barman & Yates, 2017). The BLA and CM are crucial for emotional and social experiences (Izquierdo et al., 2016;Rosenberger et al., 2019), and amygdala subnucleus dysfunction has been linked to several psychiatric disorders (Lowe et al., 2015;Ressler et al., 2011). Furthermore, studies have revealed that BLA structural alternation is associated with pain sensitivity (Zhang et al., 2021), and subnuclei volume alteration has been found in patients with psychiatric disorder (Armio et al., 2020;Zhang et al., 2020). Nevertheless, clinical experience with BLA deep brain stimulation (DBS) results in both positive and negative responses (Avecillas-Chasin et al., 2020;. One major issue with varying outcomes may be caused primarily by the complex role of the human amygdala subnuclei. Thus, defining the functions of amygdala subnuclei is fundamental to the amygdala DBS clinical practice. The explicit role of the human amygdala subnuclei remains a subject of debate, as different data indicate different roles at the subnuclei level (Seguin et al., 2021;Yoder et al., 2015). Direct electrical stimulation is a perturbation technique that allows for more unambiguous brain function delineation than fMRI, diffusion tractography, and pharmacological investigation. Direct stimulation produces simple effects in unimodal brain networks while eliciting heterogeneous and complex responses in heteromodal and transmodal networks. This study aimed to provide a comprehensive understanding of the roles of amygdala subnuclei in the human brain by direct electrical stimulation of the amygdala subnuclei.

| Patients
We performed a descriptive observational study that, retrospectively, included patients with focal drug-resistant epilepsy who underwent stereoelectroencephalography (SEEG) exploration in the Department of Neurosurgery, Xuanwu Hospital, Beijing, between 2016 and 2021 to localize the epileptogenic zone before surgery. The SEEG targets were chosen after a multidisciplinary presurgical evaluation. The study included patients with at least one intracerebral electrode implanted in the amygdala, whereas those with amygdala lesions were excluded. Furthermore, because the study focused on the clinical observations elicited by electrical stimulation, any diagnosis of psychiatric disorder or learning disability was excluded because these conditions could influence the evoked subjective perpetual and behavioral phenomena.
Electrical stimulation is routinely used to localize the epileptogenic zone and generate functional profiles in explored regions. All patients were fully informed of the purpose and risks of the SEEG procedure, and informed consent was obtained in accordance with the Declaration of Helsinki. The Institutional Review Board Committee of Xuanwu Hospital, Capital Medical University, China, approved this study.

| Surgical procedure and SEEG monitoring
Three-dimensional T1-weighted contrast MRI was performed preoperatively to avoid major vessel injury during the design of the SEEG implantation scheme. An epileptologist and a neurosurgeon carefully reviewed the implantation plan. During general anesthesia, multi-lead SEEG electrodes (Alcis, Besancon, France) with contacts ranging from 5 and 18, 0.8 mm in diameter, 2 mm in length, and 1.5 mm apart, were implanted under the guidance of a stereotactic robot, ROSA (Medtech, Montpellier, France). Following surgery, the patient underwent computed tomography to verify the location of each electrode and to detect any intracranial hemorrhages or other complications that occurred.
All the patients were monitored using video cameras over the long term to capture habitual clinical seizures. In addition, the intracranial EEG was recorded at 2048 Hz using a Nicolet data acquisition system (Natus, Orlando, USA). All recordings were obtained and referenced to a common contact placed subcutaneously.

| Reconstruction of SEEG electrodes and localization of contacts
The SEEG electrodes of each patient were then reconstructed within the Montreal Neurological Institute space using Lead-DBS (www.leaddbs.org) following a previously described protocol to reveal the electrode location of all patients. SPM12 (http://www.fil.ion.ucl.ac.uk/spm/ software/spm12) was used linearly to coregister postoperative CT images with preoperative MRI. Coregistration was manually controlled for each patient and refined if required. The images were then normalized into the ICBM 2009b NLIN asymmetric space using the SyN approach implemented in advanced normalization tools (http://stnava. github.io/ANTs) based on preoperative MRI. The locations of the contacts in the normalized images were checked using raw images. During the procedure, the images were examined slice-by-slice from the tip to the end of the electrode in the standard and individual spaces.

| Intracranial electrical stimulation protocol
During the SEEG studies, current-related intracranial stimulation was performed using a Nicolet Cortical Stimulator (Natus, Orlando, USA) as a routine procedure for diagnostic purposes and topographical delimitation of eloquent brain areas that should be considered in subsequent surgical procedures. High-frequency stimulation (frequency at 50 Hz, pulse duration of 300 μs, train duration of 5 s) was used in the same manner as previously reported (Caruana et al., 2018;Lanteaume et al., 2007;Qi et al., 2022;Winawer & Parvizi, 2016), with a bipolar mode of stimulation to adjacent contacts. Stimulus intensities ranged from 0.5 to 8 mA.
Patients were asked to recline, remain restful, calm, and awake before functional mapping. The patients were blinded to the time points at which the stimulation was delivered. An experienced neurologist and neurosurgeon asked the patients to describe any sensations and feelings in their own words, which were then assessed and categorized (Avecillas-Chasin et al., 2020;So & Alwaki, 2018). Subjective reports and clinical observations elicited by each stimulation were carefully recorded.
All the stimulations evoking a symptom were considered effective, except for those inducing symptoms throughout the propagation of the stimulus to extra-cerebral areas (e.g., meningeal pain or scalp paresthesia), which were considered non-effective. The effective electrical stimulations were then classified according to the presence or absence of electrical modification on the EEG, defined as after discharge (AD). The responses to effective stimulations were classified as physiological if never experienced by the patient before and without electrical modification (including clonic or tonic movements); (xi) racing thoughts; (xii) multimodal response (including complex symptom belonging to more than one of the above categories); and (xiii) unclassified response (when the effect was not possible to include in any of the above categories) (Balestrini et al., 2015;Mariani et al., 2021;Qi et al., 2022;So & Alwaki, 2018).

| Statistical analysis
Descriptive statistics of the variable data were analyzed. Quantitative variables are expressed as mean and standard deviation. The frequency distributions describe the evoked responses of each subnucleus. The chi-square goodness-of-fit test was performed for the related samples to evaluate the distribution of evoked responses between the hemispheres. Statistical significance was set at p < .05. Statistical analysis was performed using SPSS software version 28 (IBM SPSS, USA).

| Patient demographic and clinical data
Forty-eight right-handed patients (22 women and 26 men) were enrolled in the study. The mean age of the patients was 28.1 years. The epileptogenic zone defined by SEEG exploration included the amygdala in 33 (69%, 33/48) patients. Sixteen patients had electrodes implanted in the right amygdala, twenty-four with electrodes implanted in the left amygdala, and eight with electrodes implanted bilaterally in the amygdala.

| Contact distribution within the amygdala
There are 81 SEEG electrodes with 152 contacts implanted in the amygdala. Figure

| Visual hallucinations
In three patients, visual hallucinations were elicited 18 times. Amygdala contacts eliciting visual hallucinations were found distributed across the right and left cerebral hemispheres (2 vs. 7). One (11%, 1/9) of these contacts was in the LB, six (67%, 6/9) in the SF, one

| Somatosensory responses
Somatosensory responses were elicited 22 times in seven patients.

| Other symptoms
In two patients, auditory symptoms were elicited twice. The twoamygdala contacts that elicited auditory symptoms were distributed across the right and left cerebral hemispheres. The right contact eliciting veiled sounds was in the LB of patient 37, and the left contact eliciting familiar sounds was in the IF of patient 6.
In three patients, vestibular sensations were elicited five times.
The amygdala contacts eliciting vestibular sensations were distributed in the left cerebral hemisphere (four contacts). Three (75%, 3/4) of these were in the LB, and one (25%, 1/4) was in the SF. Vestibular sensations included dizziness elicited by one trial in the LB and vertigo elicited by four trials in the LB or SF.
Unclassified responses, such as vague symptoms, were elicited five times in patients 10 and 32. The amygdala contacts eliciting unclassified responses were distributed across the right and left cerebral hemispheres (one vs. two). Two contacts were in the bilateral LB, and one was in the left SF.

| Subnucleus distribution of clinical symptoms
Physiological symptoms were divided into eight categories ( Figure 6).

| DISCUSSION
This study provides a useful picture of direct electrical stimulation of human subnuclei. This direct "casual-effect" perturbation approach was used in a relatively large patient group to provide a reliable, functional map. Compared to most intracranial stimulation studies on the role of the amygdala in emotional experience, the main findings of this study revealed that the human amygdala plays a vital role beyond the exclusive processing of emotional experience. Further, it participates in the autonomic regulation and sensory functions of the body at the subnuclei level: (a) the amygdala was found not to mediate basic emotions specifically. However, it was involved in more general and intrinsic brain networks mediating multiple emotions. (b) Evidence revealed that the CM and other subnuclei were implicated in autonomic nervous system activity. (c) The amygdala is suggested to be involved in multimodal sensory processing, including somatosensory, visual, auditory, and vestibular sensory processing.
A key novel finding of this study was that direct amygdala stimulation could elicit several emotions related to unpleasantness and F I G U R E 7 Overall circuits revealed by amygdala-cortical evoked potential responses in 10 patients who have received amygdala-cortical evoked investigation. Each dot over the slices represents a site with significant response potential evoked by amygdala stimulation. A different color identifies the evoked response of each patient consistent with the color shown in the dot. The red and blue arrows indicate information flows.
pleasure in both the left and right amygdala subnuclei. To date, a point of debate in the neural representation of emotion is whether the amygdala specifically mediates basic emotions or plays a crucial role in multiple emotions (Gu et al., 2019;Hamann, 2012). A strong relationship between the human amygdala and basic emotions has been revealed in studies of intracranial amygdala stimulation. These studies revealed that amygdala stimulation elicited responses of fear, anxiety, sadness, and joy, with the right hemisphere stimulation producing negative responses but the left hemisphere producing both positive and negative emotions (Bijanki et al., 2014;Inman et al., 2020;Lanteaume et al., 2007;Meletti et al., 2006). Negative emotional responses were induced more frequently than positive emotional experiences, which is consistent with previous research.
However, the emotions of disdain, grief, and tenderness induced by direct amygdala stimulation in this study have not been previously described. In this study, negative and positive emotions were evoked in both hemispheres. This is consistent with several fMRI meta-analyses that showed no amygdala lateralization based on stimuli valence (Baas et al., 2004). BLA stimulation has been revealed to trigger fear-related responses in the amygdala subnuclei (Inman et al., 2020). In this study, the BLA was also related to happiness, tenderness, disgust, and disdain. SF stimulation is associated with fear, grief, relaxation, and anxiety. It can be assumed that the amygdala is involved in general and intrinsic brain networks mediating multiple emotions, regardless of lateralization to the left or right hemispheres.
The amygdala, anterior cingulate cortex, insular cortex, thalamus, hypothalamus, periaqueductal grey matter, parabrachial nucleus, and several medullary regions-regulate the central processing of sympathetic and parasympathetic outflow (Thijs et al., 2021). However, few studies have investigated the effects of intracranial stimulation on the ANS. Previous studies have revealed that amygdala stimulation induces apnea (Nobis et al., 2018) and bradycardia (Inman et al., 2020 In this study, one important finding was that amygdala stimulation induced olfactory, somatosensory, auditory, visual, and vestibular sensations. The SF in rodents receives inputs from the main olfactory bulb, however, not the accessory olfactory bulb, suggesting its important role in the main olfactory system (McDonald, 1998). The olfactory bulb projects directly to the SF (also called the periamygdaloid complex) in humans (Allison, 1954;Crosby & Humphrey, 1941). SF stimulation produces olfactory seizures, indicating SF involvement in the olfactory function system. In addition, whole-brain functional network analysis revealed strong connectivity between the SF and the temporal pole, fusiform cortex, hippocampus, parahippocampal gyrus, orbitofrontal cortex, entorhinal cortex, and dorsal pons (Noto et al., 2021). These Previous studies have revealed that the amygdala receives a wide range of inputs from cortical and subcortical regions, which are involved in information processing and emotion generating (Diano et al., 2016). The BLA is generally regarded as the gatekeeper for sensory inputs from the visual, auditory, somatosensory, olfactory, and vestibular systems (LeDoux, 2007). In addition, other amygdala regions receive inputs from the BLA and other brain areas, allowing the amygdala to process a wide range of information. Intracranial recordings of the human amygdala response to sensory stimuli have revealed a similar timing of auditory and visual sensory information (Dominguez-Borras et al., 2019). However, both afferent inputs into the amygdala and neuronal membrane potential may be reflected in the evoked response potential during intracranial recordings. Another important finding in this study was that amygdala stimulation induced visual, auditory, somatosensory, olfactory, and vestibular sensations.
The ACEP in our study has also revealed that the amygdala is functionally connected with the prefrontal cortex, anterior cingulate cortex, hypothalamus, fusiform cortex, entorhinal cortex, insular cortex, and thalamus, which is consistent with findings of previous studies.
This highlights the direct involvement of the amygdala in sensory processing, autonomic regulation, and emotion generation in humans.
This study had some limitations.