Faculty Bibliography

Departing from traditional linguistic models, advances in deep learning have resulted in a new type of predictive (autoregressive) deep language models (DLMs). Using a self-supervised next-word prediction task, these models generate appropriate linguistic responses in a given context. In the current study, nine participants listened to a 30-min podcast while their brain responses were recorded using electrocorticography (ECoG). We provide empirical evidence that the human brain and autoregressive DLMs share three fundamental computational principles as they process the same natural narrative: (1) both are engaged in continuous next-word prediction before word onset; (2) both match their pre-onset predictions to the incoming word to calculate post-onset surprise; (3) both rely on contextual embeddings to represent words in natural contexts. Together, our findings suggest that autoregressive DLMs provide a new and biologically feasible computational framework for studying the neural basis of language.

BACKGROUND:Tuberous sclerosis complex (TSC) and some focal cortical dysplasias (FCDs) are associated with dysfunctional mTOR signaling, resulting in increased cell growth and ribosomal S6 protein phosphorylation (phospho-S6). mTOR inhibitors can reduce TSC tumor growth and seizure frequency, and preclinical FCD studies indicate seizure suppression. This pilot study evaluated safety of mTOR inhibitor everolimus in treatment resistant (failure of >2 anti-seizure medications) TSC and FCD patients undergoing surgical resection and to assess mTOR signaling and molecular pathways. METHODS AND FINDINGS/RESULTS:We evaluated everolimus in 14 treatment resistant epilepsy patients undergoing surgical resection (4.5 mg/m2 daily for 7 days; n = 4 Active, mean age 18.3 years, range 4-26; n = 10, Control, mean age 13.1, range 3-45). Everolimus was well tolerated. Mean plasma everolimus in Active participants were in target range (12.4 ng/ml). Brain phospho-S6 was similar in Active and Control participants with a lower trend in Active participants, with Ser235/236 1.19-fold (p = 0.67) and Ser240/244 1.15-fold lower (p = 0.66). Histologically, Ser235/236 was 1.56-fold (p = 0.37) and Ser240/244 was 5.55-fold lower (p = 0.22). Brain proteomics identified 11 proteins at <15% false discovery rate associated with coagulation system (p = 1.45x10-9) and acute phase response (p = 1.23x10-6) activation. A weighted gene correlation network analysis (WGCNA) of brain proteomics and phospho-S6 identified 5 significant modules. Higher phospho-S6 correlated negatively with cellular respiration and synaptic transmission and positively with organophosphate metabolic process, nuclear mRNA catabolic process, and neuron ensheathment. Brain metabolomics identified 14 increased features in Active participants, including N-acetylaspartylglutamic acid. Plasma proteomics and cytokine analyses revealed no differences. CONCLUSIONS:Short-term everolimus before epilepsy surgery in TSC and FCD resulted in no adverse events and trending lower mTOR signaling (phospho-S6). Future studies should evaluate implications of our findings, including coagulation system activation and everolimus efficacy in FCD, in larger studies with long-term treatment to better understand molecular and clinical effects. CLINICAL TRIALS REGISTRATION/BACKGROUND:ClinicalTrials.gov NCT02451696.

One-third of epilepsy patients suffer from medication-resistant seizures. While surgery to remove epileptogenic tissue helps some patients, 30-70% of patients continue to experience seizures following resection. Surgical outcomes may be improved with more accurate localization of epileptogenic tissue. We have previously developed novel thin-film, subdural electrode arrays with hundreds of microelectrodes over a 100-1000 mm² area to enable high-resolution mapping of neural activity. Here, we used these high-density arrays to study microscale properties of human epileptiform activity. We performed intraoperative micro-electrocorticographic recordings in nine patients with epilepsy. In addition, we recorded from four patients with movement disorders undergoing deep brain stimulator implantation as non-epileptic controls. A board-certified epileptologist identified microseizures, which resembled electrographic seizures normally observed with clinical macroelectrodes. Recordings in epileptic patients had a significantly higher microseizure rate (2.01 events/min) than recordings in non-epileptic subjects (0.01 events/min; permutation test, P = 0.0068). Using spatial averaging to simulate recordings from larger electrode contacts, we found that the number of detected microseizures decreased rapidly with increasing contact diameter and decreasing contact density. In cases in which microseizures were spatially distributed across multiple channels, the approximate onset region was identified. Our results suggest that micro-electrocorticographic electrode arrays with a high density of contacts and large coverage are essential for capturing microseizures in epilepsy patients and may be beneficial for localizing epileptogenic tissue to plan surgery or target brain stimulation.

Brainstem nuclei dysfunction is implicated in sudden unexpected death in epilepsy. In animal models, deficient serotonergic activity is associated with seizure-induced respiratory arrest. In humans, glia are decreased in the ventrolateral medullary pre-Botzinger complex that modulate respiratory rhythm, as well as in the medial medullary raphe that modulate respiration and arousal. Finally, sudden unexpected death in epilepsy cases have decreased midbrain volume. To understand the potential role of brainstem nuclei in sudden unexpected death in epilepsy, we evaluated molecular signalling pathways using localized proteomics in microdissected midbrain dorsal raphe and medial medullary raphe serotonergic nuclei, as well as the ventrolateral medulla in brain tissue from epilepsy patients who died of sudden unexpected death in epilepsy and other causes in diverse epilepsy syndromes and non-epilepsy control cases (n = 15-16 cases per group/region). Compared with the dorsal raphe of non-epilepsy controls, we identified 89 proteins in non-sudden unexpected death in epilepsy and 219 proteins in sudden unexpected death in epilepsy that were differentially expressed. These proteins were associated with inhibition of EIF2 signalling (P-value of overlap = 1.29 × 10^-8, z = -2.00) in non-sudden unexpected death in epilepsy. In sudden unexpected death in epilepsy, there were 10 activated pathways (top pathway: gluconeogenesis I, P-value of overlap = 3.02 × 10^-6, z = 2.24) and 1 inhibited pathway (fatty acid beta-oxidation, P-value of overlap = 2.69 × 10^-4, z = -2.00). Comparing sudden unexpected death in epilepsy and non-sudden unexpected death in epilepsy, 10 proteins were differentially expressed, but there were no associated signalling pathways. In both medullary regions, few proteins showed significant differences in pairwise comparisons. We identified altered proteins in the raphe and ventrolateral medulla of epilepsy patients, including some differentially expressed in sudden unexpected death in epilepsy cases. Altered signalling pathways in the dorsal raphe of sudden unexpected death in epilepsy indicate a shift in cellular energy production and activation of G-protein signalling, inflammatory response, stress response and neuronal migration/outgrowth. Future studies should assess the brain proteome in relation to additional clinical variables (e.g. recent tonic-clonic seizures) and in more of the reciprocally connected cortical and subcortical regions to better understand the pathophysiology of epilepsy and sudden unexpected death in epilepsy.

Epilepsy affects more than three million people in the United States. In approximately one-third of this population, anti-seizure medications do not control seizures. Many patients pursue surgical treatment that can include a procedure involving the implantation of electrodes for intracranial monitoring of seizure activity. For these cases, accurate mapping of the implanted electrodes on a patient's brain is crucial in planning the ultimate surgical treatment. Traditionally, electrode mapping results are presented in static figures that do not allow for dynamic interactions and visualizations. In collaboration with a clinical research team at a Level 4 Epilepsy Center, we developed N-Tools-Browser, a web-based software using WebGL and the X-Toolkit (XTK), to help clinicians interactively visualize the location and functional properties of implanted intracranial electrodes in 3D. Our software allows the user to visualize the seizure focus location accurately and simultaneously display functional characteristics (e.g., results from electrical stimulation mapping). Different visualization modes enable the analysis of multiple electrode groups or individual anatomical locations. We deployed a prototype of N-Tools-Browser for our collaborators at the New York University Grossman School of Medicine Comprehensive Epilepsy Center. Then, we evaluated its usefulness with domain experts on clinical cases.

BACKGROUND AND OBJECTIVES:We compared heart rate variability (HRV) in sudden unexpected death in epilepsy (SUDEP) cases and living epilepsy controls. METHODS:This international, multicenter, retrospective, nested case-control study examined patients admitted for video-EEG monitoring (VEM) between January 1, 2003, and December 31, 2014, and subsequently died of SUDEP. Time domain and frequency domain components were extracted from 5-minute interictal ECG recordings during sleep and wakefulness from SUDEP cases and controls. RESULTS:= 0.209). CONCLUSIONS:Reduced short-term LFP, which is a validated biomarker for sudden death, was associated with SUDEP. Increased HFP was associated with longer survival and may be cardioprotective in SUDEP. HRV quantification may help stratify individual SUDEP risk. CLASSIFICATION OF EVIDENCE:This study provides Class III evidence that in patients with epilepsy, some measures of HRV are associated with SUDEP.

Perception results from the interplay of sensory input and prior knowledge. Despite behavioral evidence that long-term priors powerfully shape perception, the neural mechanisms underlying these interactions remain poorly understood. We obtained direct cortical recordings in neurosurgical patients as they viewed ambiguous images that elicit constant perceptual switching. We observe top-down influences from the temporal to occipital cortex, during the preferred percept that is congruent with the long-term prior. By contrast, stronger feedforward drive is observed during the non-preferred percept, consistent with a prediction error signal. A computational model based on hierarchical predictive coding and attractor networks reproduces all key experimental findings. These results suggest a pattern of large-scale information flow change underlying long-term priors' influence on perception and provide constraints on theories about long-term priors' influence on perception.

Humans form lasting memories of stimuli that were only encountered once. This naturally occurs when listening to a story, however it remains unclear how and when memories are stored and retrieved during story-listening. Here, we first confirm in behavioral experiments that participants can learn about the structure of a story after a single exposure and are able to recall upcoming words when the story is presented again. We then track mnemonic information in high frequency activity (70-200 Hz) as patients undergoing electrocorticographic recordings listen twice to the same story. We demonstrate predictive recall of upcoming information through neural responses in auditory processing regions. This neural measure correlates with behavioral measures of event segmentation and learning. Event boundaries are linked to information flow from cortex to hippocampus. When listening for a second time, information flow from hippocampus to cortex precedes moments of predictive recall. These results provide insight on a fine-grained temporal scale into how episodic memory encoding and retrieval work under naturalistic conditions.

OBJECTIVE:Brain functions such as perception, motor control, learning, and memory arise from the coordinated activity of neuronal assemblies distributed across multiple brain regions. While major progress has been made in understanding the function of individual neurons, circuit interactions remain poorly understood. A fundamental obstacle to deciphering circuit interactions is the limited availability of research tools to observe and manipulate the activity of large, distributed neuronal populations in humans. Here we describe the development, validation, and dissemination of flexible, high-resolution, thin-film (TF) electrodes for recording neural activity in animals and humans. APPROACH/METHODS:We leveraged standard flexible printed-circuit manufacturing processes to build high-resolution TF electrode arrays. We used biocompatible materials to form the substrate (liquid crystal polymer; LCP), metals (Au, PtIr, and Pd), molding (medical-grade silicone), and 3D-printed housing (nylon). We designed a custom, miniaturized, digitizing headstage to reduce the number of cables required to connect to the acquisition system and reduce the distance between the electrodes and the amplifiers. A custom mechanical system enabled the electrodes and headstages to be pre-assembled prior to sterilization, minimizing the setup time required in the operating room. PtIr electrode coatings lowered impedance and enabled stimulation. High-volume, commercial manufacturing enables cost-effective production of LCP-TF electrodes in large quantities. MAIN RESULTS/RESULTS:Our LCP-TF arrays achieve 25× higher electrode density, 20× higher channel count, and 11× reduced stiffness than conventional clinical electrodes. We validated our LCP-TF electrodes in multiple human intraoperative recording sessions and have disseminated this technology to >10 research groups. Using these arrays, we have observed high-frequency neural activity with sub-millimeter resolution. SIGNIFICANCE/CONCLUSIONS:Our LCP-TF electrodes will advance human neuroscience research and improve clinical care by enabling broad access to transformative, high-resolution electrode arrays.

OBJECTIVE:To identify the molecular signaling pathways underlying sudden unexpected death in epilepsy (SUDEP) and high-risk SUDEP compared to epilepsy control patients. METHODS:For proteomics analyses, we evaluated the hippocampus and frontal cortex from microdissected post-mortem brain tissue of 12 SUDEP and 14 non-SUDEP epilepsy patients. For transcriptomics analyses, we evaluated hippocampus and temporal cortex surgical brain tissue from mesial temporal lobe epilepsy (MTLE) patients: 6 low-risk and 8 high-risk SUDEP as determined by a short (< 50 seconds) or prolonged (≥ 50 seconds) postictal generalized EEG suppression (PGES) that may indicate severely depressed brain activity impairing respiration, arousal, and protective reflexes. RESULTS:In autopsy hippocampus and cortex, we observed no proteomic differences between SUDEP and non-SUDEP epilepsy patients, contrasting with our previously reported robust differences between epilepsy and non-epilepsy control patients. Transcriptomics in hippocampus and cortex from surgical epilepsy patients segregated by PGES identified 55 differentially expressed genes (37 protein-coding, 15 lncRNAs, three pending) in hippocampus. CONCLUSION/CONCLUSIONS:The SUDEP proteome and high-risk SUDEP transcriptome were similar to other epilepsy patients in hippocampus and frontal cortex, consistent with diverse epilepsy syndromes and comorbidities associated with SUDEP. Studies with larger cohorts and different epilepsy syndromes, as well as additional anatomic regions may identify molecular mechanisms of SUDEP.