Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Cellular-Level Organization of Ventral Sensorimotor Cortex During Fluent Speech
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Quinn R. Greicius1, Tommy Hosman1, Tanay Poddar1, Ethan J. Kato1, Yuhan Hao1, Jessie R. Liu1, Duo Xu1, Matthew K. Leonard1, Barundeb Dutta2, Marleen Welkenhuysen2, Jason E. Chung1, Jacob S. Young1, Shawn L. Hervey-Jumper1, Edward F. Chang1; 1University of California San Francisco, 2IMEC, Leuven, Belgium
Understanding how the human brain produces speech at the cellular level remains a fundamental open question. Mesoscopic electrocorticography (ECoG) studies of speech production have established a somatotopic organization in the ventral sensorimotor cortex (vSMC) and suggest that articulatory kinematics, rather than abstract phonemes, are the dominant representational scheme (Bouchard 2013, Chartier 2018). Whether this featural encoding holds at the level of individual neurons in fluent speakers, and how it is organized across cell types and cortical columns, is unknown. We addressed these questions using high-density Neuropixels recordings of more than 5,000 speech-responsive single units from 31 insertions in the ventral precentral (vPrCG) and postcentral (vPoCG) gyri of 10 awake neurosurgical participants. Participants performed an isolated consonant-vowel (CV) syllable production task spanning the English consonant repertoire, as well as a fluent sentence production task. In the CV task, single neurons were tuned to articulatory-phonetic features rather than individual phonemes. We identified abundant cellular selectivity for labial, alveolar, and velar places of articulation, as well as voicing, nasality, and manner of articulation. Population geometry at each site recapitulated the articulatory-phonetic axes seen at the single unit level, with significant clustering by place, manner, and voicing, and per-site clustering strength differing by feature. Similarly, during fluent speech, neuronal activity was better captured by articulatory kinematic trajectories than by any individual phoneme. Vowel feature encoding was also identified at the single-neuron level in fluent speech, with tuning profiles in formant space that reflected the articulatory categories of height and centrality. Across the neuronal population sampled by a given probe in vSMC, the distribution of single-neuron tuning profiles exhibited significant preference for certain articulatory categories. In the local populations where such preferences were most pronounced, single-unit tuning was nearly uniform. More generally, however, these preferences emerged despite considerable tuning heterogeneity. Similar preferences were observed among probes at adjacent insertion sites, reflecting the local consistency of the spatial organization previously shown in ECoG. This novel dataset reveals a high-resolution cellular architecture for speech in which articulatory features are the fundamental unit of single-neuron representations, with local tuning preferences in the context of variable degrees of heterogeneity. These findings provide a cellular foundation for understanding fluent speech production and its disruption in motor speech disorders.
Topic Areas: Speech Motor Control,