Single Neurons in Prefrontal Cortex Encode Semantic Content During Language Comprehension. (16th November 2020)

• Record Type: Journal Article
• Authors: Jamali, Mohsen; Grannan, Benjamin L; Cai, Jing; Lee, Dan; Khanna, Arjun; Fedorenko, Evelina; Williams, Ziv
• Is Part Of: Neurosurgery. Volume 67(2010)Supplement 1
• Publication Date: 2020-11-16
• Subjects: Nervous system -- Surgery -- Periodicals; 617.48005
• DOI: 10.1093/neuros/nyaa447_607
• Languages: English
• ISSNs: 0148-396X

Abstract

Abstract: INTRODUCTION: From basic sequences of speech sounds or strings of letters, humans are able to extract extraordinarily rich and complex meaning through language. This process has been shown to recruit the prefrontal cortex among other associative areas in imaging studies. The role of single neuron firing patterns in assigning meaning to linguistic auditory input remains unknown. METHODS: We performed acute single unit recordings from the posterior middle frontal gyrus of the prefrontal cortex in 10 right-handed patients undergoing awake DBS surgery. Patients listened to prerecorded 8-word sentences and non-ordered word lists that were designed to sample broadly across different semantic themes. Study words were clustered into 9 semantic domains using a word-embedding (i.e., word2vec) and dimensionality reduction approach. Single-neuron responses that were selective for certain semantic domains were detected by comparing firing rates in response to words from one domain with responses to all other words. Population decoding was performed with a multinomial logistic regression model (60% training data, 40% test data) across 1000 iterations. RESULTS: By tracking the action potential dynamics of neurons as they evolved during sentence comprehensions, we discovered that 22% of the cells in the population (48/220) were selective for at least one semantic domain ( P < .025) over sub-second timescales. These neurons exhibited responses selective to specific semantic domain(s) when presented with words in complete, natural sentences, but not when presented with non-ordered random word lists or with homonym words with identical phonetics but different semantic meanings. Using the population firing rate to predict the semantic domain of each word, the decoding accuracy reached ∼45% (compared to 11% accuracy in after bootstrapped randomization of semantic labels). Interestingly, decoding performance dropped to chance when listening to the same words but in a randomly ordered word lists. CONCLUSION: During language comprehension, dominant prefrontal neurons demonstrate semantically selective activations during language processing. This activity was only present when words occurred in a sentence, suggesting that single prefrontal neurons are involved in context-dependent comprehension of linguistic input.