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.

Importance/UNASSIGNED:Non-communicable chronic diseases (NCDs) such as obesity, type 2 diabetes, heart disease, and cancer were rare among non-western populations with traditional diets and lifestyles. As populations transitioned toward industrialized diets and lifestyles, NCDs developed. Objective/UNASSIGNED:We performed a systematic literature review to examine the effects of diet and lifestyle transitions on NCDs. Evidence Review/UNASSIGNED:We identified 22 populations that underwent a nutrition transition, eleven of which had sufficient data. Of these, we chose four populations with diverse geographies, diets and lifestyles who underwent a dietary and lifestyle transition and explored the relationship between dietary changes and health outcomes. We excluded populations with features overlapping with selected populations or with complicating factors such as inadequate data, subgroups, and different study methodologies over different periods. The selected populations were Yemenite Jews, Tokelauans, Tanushimaru Japanese, and Maasai. We also review transition data from seven excluded populations (Pima, Navajo, Aboriginal Australians, South African Natal Indians and Zulu speakers, Inuit, and Hadza) to assess for bias. Findings/UNASSIGNED:The three groups that replaced saturated fats (SFA) from animal (Yemenite Jews, Maasai) or plants (Tokelau) with refined carbohydrates had negative health outcomes (e.g., increased obesity, diabetes, heart disease). Yemenites reduced SFA consumption by >40% post-transition but men's BMI increased 19% and diabetes increased ~40-fold. Tokelauans reduced fat, dramatically reduced SFA, and increased sugar intake: obesity and diabetes rose. The Tanushimaruans transitioned to more fats and less carbohydrates and used more anti-hypertensive medications; stroke and breast cancer declined while heart disease was stable. The Maasai transitioned to lower fat, SFA and higher carbohydrates and had increased BMI and diabetes. Similar patterns were observed in the seven other populations. Conclusion/UNASSIGNED:The nutrient category most strongly associated with negative health outcomes - especially obesity and diabetes - was sugar (increased 600-650% in Yemenite Jews and Tokelauans) and refined carbohydrates (among Maasai, total carbohydrates increased 39% in men and 362% in women), while increased calories was less strongly associated with these disorders. Across 11 populations, NCDs were associated with increased refined carbohydrates more than increased calories, reduced activity or other factors, but cannot be attributed to SFA or total fat consumption.

OBJECTIVE:A recent FDA warning concerning an arrhythmogenic potential of lamotrigine created concern in the neurological community. This warning was based on in vitro studies, but no clinically relevant risk was considered. This rapid systematic review aims to elucidate the risk of lamotrigine on sudden death or electrocardiogram abnormalities. METHODS:We conducted a systematic search of Ovid Medline and Ovid Embase, including randomized controlled trials and observational studies, studies of people with or without epilepsy, with one of the following outcome measures: SUDEP and sudden cardiac death, as well as the development or worsening of electrocardiogram abnormalities. All titles and abstracts were independently screened, and the full texts of relevant studies were obtained. We re-evaluated the sudden death definitions used in all included studies, as some could have used unclear or overlapping definitions. We used the American Academy of Neurology risk of bias tool to evaluate the class of evidence and the GRADE approach to evaluate our confidence in the evidence. RESULTS:We included 26 studies with 24,962 participants, of whom 2,326 used lamotrigine. Twelve studies showed no significant risk of SUDEP for lamotrigine users. One study reporting on sudden cardiac death and three studies with unclear sudden death definitions did not report an elevated risk of death in lamotrigine users compared to controls. In 10 studies reporting on electrocardiogram parameters, there was no statistically significant increased risk among lamotrigine users except for two studies. These two studies reported either "slight increases" in PR interval or an increased PQ interval that the primary study authors felt to be more related to structural cardiac differences rather than an effect of lamotrigine. One study was rated class II while all others were class III or IV. We had "very low confidence" in the evidence following the GRADE assessment. None of the studies examined the risk of lamotrigine in people with pre-existing cardiac conditions. CONCLUSION/CONCLUSIONS:There is insufficient evidence to support or refute that lamotrigine is associated with sudden death or electrocardiogram changes, in people with or without epilepsy as compared to ASM or placebo. This is due to the high risk of bias in most studies and low precision and inconsistency in the reported results.

OBJECTIVES/OBJECTIVE:Medical cannabis formulations with cannabidiol (CBD) and delta-9-tetrahydrocannabinol (THC) are widely used to treat epilepsy. We studied the safety and efficacy of two formulations. METHODS:We prospectively observed 29 subjects (12 to 46 years old) with treatment-resistant epilepsies (11 Lennox-Gastaut syndrome; 15 with focal or multifocal epilepsy; three generalized epilepsy) were treated with medical cannabis (1THC:20CBD and/or 1THC:50CBD; maximum of 6 mg THC/day) for ≥24 weeks. The primary outcome was change in convulsive seizure frequency from the pre-treatment baseline to the stable optimal dose phase. RESULTS:There were no significant differences during treatment on stable maximal doses for convulsive seizure frequency, seizure duration, postictal duration, or use of rescue medications compared to baseline. No benefits were seen for behavioral disorders or sleep duration; there was a trend for more frequent bowel movements compared to baseline. Ten adverse events occurred in 6/29 patients, all were transient and most unrelated to study medication. No serious adverse events were related to study medication. INTERPRETATION/CONCLUSIONS:Our prospective observational study of two high-CBD/low-THC formulations found no evidence of efficacy in reducing seizures, seizure duration, postictal duration, or rescue medication use. Behavioral disorders or sleep duration was unchanged. Study medication was generally well tolerated. The doses of CBD used were lower than prior studies. Randomized trials with larger cohorts are needed, but we found no evidence of efficacy for two CBD:THC products in treating epilepsy, sleep, or behavior in our population.

OBJECTIVE:To estimate the timing of cannabidiol (CBD) treatment effect (seizure reduction and adverse events [AEs]) onset, we conducted a post hoc analysis of GWPCARE6 (NCT02544763), a randomized, placebo-controlled, phase 3 trial in patients with drug-resistant epilepsy associated with tuberous sclerosis complex (TSC). METHODS:; 100 mg/mL oral solution) at 25 mg/kg/day (CBD25) or 50 mg/kg/day (CBD50) or placebo for 16 weeks (4-week titration, 12-week maintenance). Treatment started at 5 mg/kg/day for all groups and reached 25 mg/kg/day on Day 9 and 50 mg/kg/day on Day 29. Percentage change from baseline in TSC-associated seizure (countable focal or generalized) count was calculated by cumulative day (i.e., including all previous days). Time to onset and resolution of AEs were evaluated. RESULTS:Of 224 patients, 75 were randomized to CBD25, 73 to CBD50, and 76 to placebo. Median (range) age was 11.3 (1.1-56.8) years. Patients had discontinued a median (range) of 4 (0-15) antiseizure medications and were currently taking 3 (0-5). Difference in seizure reduction between CBD and placebo emerged on Day 6 (titrated dose, 15 mg/kg/day) and became nominally significant (p<0.049) by Day 10. Separation between placebo and CBD in ≥50% responder rate also emerged by Day 10. Onset of AEs occurred during the first 2 weeks of the titration period in 61% of patients (CBD25, 61%; CBD50, 67%; placebo, 54%). In patients with an AE, resolution occurred within 4 weeks of onset in 42% of placebo and 27% of CBD patients and by end of trial in 78% of placebo and 51% of CBD patients. SIGNIFICANCE/CONCLUSIONS:Onset of treatment effect occurred within 6-10 days. AEs lasted longer for CBD than placebo but the most common (diarrhea, decreased appetite, and somnolence) resolved during the 16-week trial in most patients.

To derive meaning from sound, the brain must integrate information across many timescales. What computations underlie multiscale integration in human auditory cortex? Evidence suggests that auditory cortex analyses sound using both generic acoustic representations (for example, spectrotemporal modulation tuning) and category-specific computations, but the timescales over which these putatively distinct computations integrate remain unclear. To answer this question, we developed a general method to estimate sensory integration windows-the time window when stimuli alter the neural response-and applied our method to intracranial recordings from neurosurgical patients. We show that human auditory cortex integrates hierarchically across diverse timescales spanning from ~50 to 400 ms. Moreover, we find that neural populations with short and long integration windows exhibit distinct functional properties: short-integration electrodes (less than ~200 ms) show prominent spectrotemporal modulation selectivity, while long-integration electrodes (greater than ~200 ms) show prominent category selectivity. These findings reveal how multiscale integration organizes auditory computation in the human brain.

Hearing one's own voice is critical for fluent speech production as it allows for the detection and correction of vocalization errors in real time. This behavior known as the auditory feedback control of speech is impaired in various neurological disorders ranging from stuttering to aphasia; however, the underlying neural mechanisms are still poorly understood. Computational models of speech motor control suggest that, during speech production, the brain uses an efference copy of the motor command to generate an internal estimate of the speech output. When actual feedback differs from this internal estimate, an error signal is generated to correct the internal estimate and update necessary motor commands to produce intended speech. We were able to localize the auditory error signal using electrocorticographic recordings from neurosurgical participants during a delayed auditory feedback (DAF) paradigm. In this task, participants hear their voice with a time delay as they produced words and sentences (similar to an echo on a conference call), which is well known to disrupt fluency by causing slow and stutter-like speech in humans. We observed a significant response enhancement in auditory cortex that scaled with the duration of feedback delay, indicating an auditory speech error signal. Immediately following auditory cortex, dorsal precentral gyrus (dPreCG), a region that has not been implicated in auditory feedback processing before, exhibited a markedly similar response enhancement, suggesting a tight coupling between the 2 regions. Critically, response enhancement in dPreCG occurred only during articulation of long utterances due to a continuous mismatch between produced speech and reafferent feedback. These results suggest that dPreCG plays an essential role in processing auditory error signals during speech production to maintain fluency.

Reconstructing intended speech from neural activity using brain-computer interfaces holds great promises for people with severe speech production deficits. While decoding overt speech has progressed, decoding imagined speech has met limited success, mainly because the associated neural signals are weak and variable compared to overt speech, hence difficult to decode by learning algorithms. We obtained three electrocorticography datasets from 13 patients, with electrodes implanted for epilepsy evaluation, who performed overt and imagined speech production tasks. Based on recent theories of speech neural processing, we extracted consistent and specific neural features usable for future brain computer interfaces, and assessed their performance to discriminate speech items in articulatory, phonetic, and vocalic representation spaces. While high-frequency activity provided the best signal for overt speech, both low- and higher-frequency power and local cross-frequency contributed to imagined speech decoding, in particular in phonetic and vocalic, i.e. perceptual, spaces. These findings show that low-frequency power and cross-frequency dynamics contain key information for imagined speech decoding.

OBJECTIVE:The number, unpredictability, and severity of seizures experienced by patients with Dravet syndrome (DS) negatively impact quality of life (QOL) for patients, caregivers, and families. Metrics are needed to assess whether patients with residual seizures have moved meaningfully toward seizure freedom after treatment with new antiseizure medications. METHODS:We evaluated the time required postrandomization for each patient to experience the same number of seizures experienced during baseline (i.e., time-to-nth seizure), using a post hoc time-to-event (TTE) analysis of data from two Phase 3 placebo-controlled trials of adjunctive fenfluramine for DS (Study 1, N = 119; Study 2, N = 87). Patients aged 2-19 years were randomized to placebo or adjunctive fenfluramine (Study 1: .7 mg/kg/day or .2 mg/kg/day; Study 2: .4 mg/kg/day with stiripentol). Data were analyzed by Kaplan-Meier TTE curves and waterfall plots. RESULTS:The proportion of patients who never reached baseline seizure frequency was greater with fenfluramine than with placebo (Study 1: fenfluramine .7 mg/kg/day, 60%; fenfluramine .2 mg/kg/day, 31%; placebo, 13%; Study 2: fenfluramine .4 mg/kg/day, 58%; placebo, 2%). Median time-to-nth seizure was longer after fenfluramine than after placebo (Study 1: fenfluramine .7 mg/kg/day, 13 weeks; .2 mg/kg/day, 10 weeks; placebo, 7 weeks; Study 2: fenfluramine .4 mg/kg/day, 13 weeks; placebo, 5 weeks; p < .001). Longest duration of convulsive seizure-free days was increased in active groups versus the placebo group (Study 1: fenfluramine .7 and .2 mg/kg/day, 25.0 and 15.0 days; placebo, 9.5 days [p = .0001; p = .0352]; Study 2: fenfluramine .4 mg/kg/day, 22.0 days; placebo, 13.0 days [p = .004]). The most common adverse events included decreased appetite, pyrexia, upper respiratory tract infection, diarrhea, and fatigue. SIGNIFICANCE/CONCLUSIONS:These data demonstrate that fenfluramine can significantly reduce day-to-day seizure burden in patients with DS, providing prolonged periods of convulsive seizure-free days, which may help reduce the physical and emotional disease toll while improving health-related QOL for patients and caregivers.