Poster Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions

Cortical dynamics are predictive of grunt type in minipigs

Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai

Scott Greenhorn¹, Laure Gicquel¹, Marie Lion¹, Mehrdad Koshnevis¹, Anne Quesnel-Hellman¹, Julia Syrett¹, Samuel Carlier¹, Clément Hébert¹, Florian Falleger², Stéphanie Lacour², Lionel Rousseau³, Blaise Yvert¹; ¹Grenoble Institute for Neurosciences, Univ. Grenoble Alpes, INSERM, Grenoble, FR, ²Laboratory for Soft Bioelectronic Interfaces, EPFL, Geneva, CH, ³ESYCOM-ESIEE, Univ. Gustave Eiffel, Noisy-le-Grand, FR

INTRODUCTION Control of speech in humans is underlied by widely distributed cortical networks. Speech production is due to cognitive processes lasting hundreds of milliseconds before vocal onset [1]. While distributed cortical networks have been identified underlying vocal production in a variety of nonhuman mammals and birds [2], vocal planning in non-human species remains largely under-explored. We investigate cortical dynamics in minipigs, social and easily trainable animals with a large vocal repertoire [3]. We attempt to demonstrate that large cortical networks underlie minipig vocalizations and are engaged long before vocal onset, with activity predicting vocal behavior. METHODS Two minipigs were chronically implanted with electrocorticography arrays in the left hemisphere, including motor and auditory areas. A grunt recording from one of the pig’s congeners was played back repeatedly to the animal to stimulate vocal responses, during which simultaneous audio and neural data was recorded. Grunts were labelled as either “long” or “short”, based on low-excitement grunts identified by [3], with other noises removed. Neural data was considered starting 1.2 s before each grunt onset, filtered into local field potential (<10 Hz) and 7 frequency bands at 100 Hz. Partial Least Squares reduced the feature dimensionality before applying a linear classifier at each time step to identify the time ranges predictive of grunt type. The procedure was applied 100 times, obtaining an average accuracy for each time step. For reference, the same procedure was applied with shuffled labels to estimate the decoding accuracy expected by chance. The recorded grunts were labelled and played to three pigs over three sessions. The response rate for short and long type playbacks were calculated and compared to a chance distribution. RESULTS The grunt type could be predicted from the cortical activity up to 700 ms before the grunt onset with up to 70% accuracy. Significant (Wilcoxon ranked test) peaks in prediction accuracy are visible in the auditory cortex near 600 ms before grunt onset, and in the motor cortex near 200 ms before grunt onset for both pigs. After playing back the recorded grunts to other pigs, a consistent trend of higher response rates (p-value 0.0111, 0.0002, 0.0412; Fisher exact test) and faster responses to long playbacks was observed. CONCLUSION We show that cortical dynamics in the auditory followed by motor cortex predict the production of long vs. short grunts in minipigs, compatible with a vocal plan preceding vocal onset by about 600 ms. The response rates to short and long grunts were statistically significantly different for all pigs, indicating that the grunt type meaningfully influences congeners’ behavior. REFERENCES [1] Indefrey, P. & Levelt, W. J. M. Levelt. (2004) The Spatial and Temporal Signatures of Word Production Components. Cognition 92, nᵒˢ 1‑2. [2] Nieder, A & Mooney, R. (2020) The Neurobiology of Innate, Volitional and Learned Vocalizations in Mammals and Birds. Philosophical Transactions of the Royal Society B: Biological Sciences 375, nᵒ 1789: 20190054. [3] Kiley, M. (1972) The Vocalizations of Ungulates, Their Causation and Function. Zeitschrift Für Tierpsychologie 31, nᵒ 2 (pp. 171‑222).

Topic Areas: Animal Communication and Comparative/Evolutionary Studies,