3D directional tuning in the orofacial sensorimotor cortex during natural feeding and drinking

DirecƟonal tongue movements are crucial for feeding and speech, ensuring proper food posiƟoning for chewing and swallowing, as well as accurate sound producƟon. While direcƟonal tuning in the arm region of the sensorimotor cortex during reaching tasks is well-studied, liƩ le is known about how 3D tongue direc Ɵon is encoded in the orofacial sensorimotor cortex (OSMCx) during natural behaviors. Understanding this neural representa Ɵon has important implica Ɵons for rehabilitaƟng individuals with orolingual dysfunc Ɵons. This study examines the direcƟonal tuning and populaƟon dynami cs in OSMCx during naturalisƟc feeding and drinking, and how these are affected by sensory loss. Using biplanar videoradiography, we tracked implanted tongue markers in behaving rhesus macaques ( Macaca mula Ʃ a) and simultaneously recorded 3D posiƟonal data with spiking acƟvity from chronically implanted microelectrode arrays in primary motor (MIo) and somatosensory (SIo) areas of the orofacial cortex. In some sessions, tasks were preceded by bilateral nerve block injec Ɵons to the sensory branches of the trigeminal nerve. ModulaƟon to 3D tongue direcƟon during feeding and drinking was found in most MIo and SIo neurons. DirecƟonal informaƟon in both individual- and populaƟon-level was higher in feeding and was more robust in MIo. Following sensory loss, alteraƟons in tongue kinema Ɵcs were accompanied by changes in direcƟonal informaƟon in MIo and SIo, manifesƟng as modificaƟons in both individual neuron tuning characteris Ɵcs and the broader dynamics of popula Ɵon-level neural ac Ɵvity. Overall, this study advances our understanding of how OSMCx contributes to complex, coordinated control of naturalis Ɵc tongue movements. It expands our current knowledge of orofacial control to three dimensions and demonstrates the specificity and adaptability of populaƟon acƟvity in MIo and SIo in response to different behavioral contexts. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint IntroducƟ on Motor and somatosensory cor Ɵcal neurons modulate their spiking ac Ɵvity based on movement direcƟon as seen in arm reaching tasks (Georgopoulos et al., 1988; Schwartz et al., 1988a; Prud’homme and Kalaska, 1994) and orofacial behaviors. In the primary motor (MIo) and primary somatosensory (SIo) areas of the orofacial sensorimotor cortex (OSMCx), neurons encode the direcƟon of voluntary tongue protrusion (Murray and Sessle, 1992; Lin et al., 1994a) and semi- automaƟc tongue movements in chewing and swallowing (Sessle et al., 2005b). Extensive research has explored how the arm region of the sensorimotor cortex encodes movement direcƟon (Ajemian et al., 2000; Georgopoulos et al., 2007; Churchland et al., 2012; Lillicrap and ScoƩ , 2013). Since the tongue is enclosed within the oral cavity and thus hidden from view, it has proved difficult to study the neuromechanical processes underlying direc Ɵonal tongue movements that are essenƟal for these behaviors (Hiiemae and Palmer, 2003). Thus, considerably less is known about how 3D tongue direc Ɵon is encoded in the OSMCx and the role of tac Ɵle sensaƟon (Bach-y-Rita et al., 1998; Lamm et al., 2005; Lozano et al., 2009) during natural feeding and drinking. This knowledge has important implicaƟons for improving evaluaƟon and treatment strategies for individuals with sensorimotor dysfuncƟons (Takizawa et al., 2016; Avivi-Arber and Sessle, 2017). The OSMCx plays an important role in the coordinaƟon of complex tongue movements. Seminal studies on the direc Ɵonal tuning proper Ɵes of OSM Cx neurons by Sessle and colleagues employed varying locaƟons of spouts that delivered a juice reward to elicit direc Ɵonal tongue protrusions without tracking tongue movements. A later study incorporated tracking of 2D tongue movements using videofluoroscopy during voluntary direcƟonal protrusions (Arce et al., 2013), but the tongue trajectories were not used to study d irecƟonal tuning. In all these prior studies, primates have been trained to interact with a computer display to elicit a tongue protrusion to a specific direcƟon on cue. There is a knowledge gap on how spiking acƟvity in the OSMCx relates to tongue movements during natural behaviors. With the development of biplanar video-radiography (Brainerd et al., 2010), it is now possible to track these 3D tongue movements within the oral cavity at a high temporal and spaƟal resoluƟon (Montuelle et al., 2020; Feilich et al., 2021). By simultaneous recording of precise tongue movements and spiking acƟvity, we have shown recently that tongue posiƟon and shape c an be accurately decoded from OSMCx during feeding (Laurence-Chasen et al., 2023). Our study invesƟgated how OSMCx encodes and decodes tongue direc Ɵon during untrained feeding and drinking, comparing acƟvity across mul Ɵple corƟcal regions. Given the importance of oral somatosensaƟon in tongue posiƟoning and bolus control during chewing and swallowing (Smith and Cutrer, 2011), we also examined its role in sensorimotor control by selecƟvely blocking oral tacƟle sensaƟon. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint

Subjects. Experiments were performed on two adult male rhesus macaques (Macaca mulaƩ a, 9– 10 kg, ages 8 and 9 years) in the University of Chicago XROMM Facility. This sample size was chosen based on precedent in the field of non-human primate motor neuroscience. All protocols were approved by the University of Chicago Animal Care and Use Commi Ʃ ee and complied with the NaƟonal InsƟtutes of Health Guide for the Care and Use of Laboratory Animals. The subjects were seated in a standard primate chair and head-fixed to keep their head posiƟon constant during feeding and drinking trials. Each trial lasted 10 seconds. In a feeding trial, a piece of food (grape, gummy bear, pasta) of roughly the same size was presented directly to the animals’ mouth using a stylus. In a drinking trial, juice was delivered through one of three spouts posiƟoned in front of the subject (Fig. 1A). Cranial Nerve V anesthesia. For some sessions, these behavioral tasks were preceded by nerve block injecƟons (0.25% Bupivacaine HCL and Epinephrine 1:200,000, 0.25 mL/injec Ɵon site) to the sensory branches of bilateral trigeminal nerves (lingual, inferior alveolar, buccal, palaƟne) to eliminate oral tacƟle sensaƟon locally and temporarily. The nerve block was administered while the subjects were under sedaƟon, and all data were collected within 90 minutes of the nerve block. Each monkey served as its own control, with nerve block feeding data collec Ɵon sessions taking place either a day before or a day aŌ er the associated control session. Nerve block drinking data collec Ɵon was performed immediately following the control drinking session. MulƟple datasets (40-60 trials) were collected for both subjects across mulƟple days. However, due to the complex and Ɵme-consuming nature of processing integrated XROMM and neural data, one session per subject, behavior, and condiƟon was used for this study. Thus, we analyzed a total of 8 datasets. Video-radiography. Prior to data collecƟon, the animals were implanted with spherical tantalum beads (1-mm diameter) in the cranium, mandible, and the tongue, from the Ɵp to the region of the circumvallate papillae. During feeding or drinking, the movement of these markers was recorded using high- resoluƟon (200 Hz , <0.1 mm) biplanar video-radiography collected with Xcitex ProCapture version 1.0.3.6. The 3D posi Ɵonal data was obtained following the previously described X-ray ReconstrucƟon of Moving Morphology (XROMM) workflow (Laurence-Chasen et al., 2020) incorporaƟng the use of XMALab (Knörlein et al., 2016) and machine learning using DeepLabCut (Mathis et al., 2018) to reconstruct the kinema Ɵc data. The 𝑥, 𝑦, 𝑧 values of the markers were then smoothed with a 30 Hz low-pass B uƩ erworth filter and transformed into a cranial coordinate space with the origin fixed at the posterior nasal spine. Gape cycles within each feeding sequence were manually idenƟfied and categorized by cycle type (manipulaƟon, stage 1 transport, chew, stage 2 transport, or swallow). Electrophysiology. Under general anesthesia, a microelectrode array was chronically implanted in four areas of the leŌ hemisphere (Supplementary file 1, Fig. 1): rostral MIo (96-electrode Utah .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint array; electrode length: 1.5 mm), caudal MIo (32-electrode FloaƟng microelectrode array (FMA), electrode length: 3.0–4.5 mm), area 1/2 (96-electrode Utah array, electrode length: 1.0 mm), and area 3a/3b (32-electrode FMA, electrode length: 4.0–8.7 mm). The neural data was recorded using Grapevine Neural Interface Processor (Ripple Neuro, Salt Lake City, UT). Signals were amplified and bandpass filtered between 0.1 Hz and 7.5 kHz and recorded digitally (16-bit) at 30 kHz per channel. Only waveforms (1.7 ms in dura Ɵon; 48 sample Ɵme points per waveform) that crossed a threshold were stored and offline spike sorted (Offline Sorter, Plexon, Dallas, TX) to remove noise and to isolate individual neurons. Neurons recorded during control feeding sessions are the same as those previously reported on in Laurence-Chasen et al., 2023. The channel name assigned to each recorded neuron was kept consistent between control and nerve block data for comparison. Data analysis 3D kinemaƟ cs. 3D tongue kinemaƟcs were recorded simultaneously with the neural data in all behavioral sessions. All data analyses were performed in MATLAB 2022b (MathWorks , NaƟck, MA). For feeding, the instantaneous 3D direc Ɵon of the tongue Ɵp marker for every 100 ms throughout each gape cycle was calculated as: 3D angle, ϑ = tan-1(‖v1×v2‖/v1⋅v2) (1) Where 𝑣ଵ is the 𝑥, 𝑦, 𝑧 posiƟon at the start of each 100-ms interval and 𝑣ଶ is the posiƟon at the end (Fig. 1B) . These direc Ɵons were then categorized based on whether the movement was negaƟve or posi Ɵve rela Ɵve to the horizontal plane (LeŌ /Right), the sagi Ʃ al plane (Inferior/Superior), and the 𝑥 axis (Posterior/Anterior). This resulted in eight direcƟons: AntSupL, AntSupR, AntInfL, AntInfR, PostSupL, PostSupR, PostInfL, and PostInfR. An equal number of 100- ms intervals from each of these direcƟons was sampled to eliminate the possible effect of different distribuƟons of kinemaƟcs across datasets, and spike data during each interval was used for neural analysis. For comparison with the drinking task, the sign was determined relaƟve to the horizontal plane, with rightward tongue movement being posi Ɵve. This is also the plane of moƟon which has been the least studied. These leŌ -right direcƟons were categorized into six 10- bins with a total range of -30  to 30 , which encapsulated most of the observed distribuƟon of direcƟons in each subject. Lingual yaw (transverse rotaƟon) and pitch (elevaƟon/depression) were also calculated to compare tuning across the lateral and verƟcal components of tongue direcƟon (Supplementary file 2, EquaƟon 2). For drinking, the direcƟon was determined by which of the three spouts juice was dispensed from during each lick. The spiking acƟvity used for neural analysis of the drinking task was from intervals of ±250 ms around each minimum protrusion of the tongue. As 100 ms was not sufficient to capture the full range of tongue moƟon during each drinking cycle, the length of Ɵme used was increased to allow a clear dis ƟncƟon between the three direcƟons. The period of 250 ms spans about 75% of the average lick length from minimum to maximum protrusion of the tongue. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint KinemaƟc performance for feeding was determined by the spread of tongue direcƟons observed across trials. For drinking trials, performance was determined by the variance of endpoint posiƟons as well as by the propor Ɵon of “failed” cycles, where the mon key missed the correct spout loca Ɵon with their tongue Ɵp. The difference between control and nerve block performance was evaluated using a two-tailed t-test and f-test. DirecƟ onal tuning of single neurons. Tongue direcƟons were subsequently compared with the firing rates of individual neurons across corƟcal areas. To determine if neurons were direcƟonally modulated, we used a bootstrap procedure (Arce et al., 2010): we resampled the firing rates from an equal number of trials in each direc Ɵon with replacement 1000 Ɵmes and computed 95% confidence intervals from the resulƟng distribuƟon to test whether the mean ranks are the same across direcƟons. The proporƟons of neurons that were found to be direc Ɵonally tuned were compared across groups using a chi-square test. Due to limited neuron counts in some cor Ɵcal regions, we combined rM1 and cM1 recordings as MIo, and areas 1/2 and 3a/3b as SIo for subsequent analyses (Supplementary file 1, Table 1). Then, mulƟple linear regression was used to determine if the firing of each neuron fit the cosine tuning func Ɵon that has been previously described for the arm area of the motor cortex (Schwartz et al., 1988b). To accomplish this, the direcƟonal components of a unit vector represen Ɵng each group of direc Ɵons were calculated. For neurons that fit the tuning func Ɵon, a preferred direcƟon (PD) in 3D space was es Ɵmated. These PDs are distributed around a unit sphere , with the origin represen Ɵng the start of the movement. The direcƟonal index was calculated as a measure of the depth of direcƟonal tuning. To determine PDs for the drinking task, we resampled the original distribuƟon of firing rates with replacement for each direc Ɵon and calculated the direc Ɵon for which a neuron exhibited its maximal firing rate over 1000 bootstrap samples. Similarly, a PD across the le Ō -right feeding direcƟons was determined for comparison. Circular concentra Ɵon (k -test) to compare distribuƟons of PDs during feeding and polar plot genera Ɵon were performed using the CircStat MATLAB toolbox (Berens, 2009). For drinking, distribu Ɵons of PDs were compared using a chi- square test. We analyzed the trial-by-trial variability of neuronal ac Ɵvity using the Fano factor, which was computed as the spike-count variance divided by spike-count mean within each session. The Fano Factor was calculated separately for each subject, task, and corƟcal region. For analysis across sessions, we used the mean-matched Fano factor (Churchland et al., 2010). Factor Analysis of popula Ɵ on ac Ɵ vity. We used Factor Analysis (FA), a linear dimensionality reducƟon method, to obtain latent trajectories of spiking ac Ɵvity and compare popula Ɵon responses to different direcƟons across trials. FA defined as: y ∼ N (μ, CC’ + R) (2) where y is the spike counts from n neurons, μ is the mean spike counts from n neurons, C (m × n) is the loading matrix mapping m latent factors to the spike counts of n neurons, and R (n × n) .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint represents the unexplained variance of independent neurons(Santhanam et al., 2009; Horrocks et al., 2024). The parameters μ, C and R were esƟmated with expecta Ɵon-maximizaƟon using DataHigh Matlab toolbox (Cowley et al., 2013). The latent factor space obtained using FA represents the shared popula Ɵon variance of neurons. To obtain latent factors of shared populaƟon acƟvity, we binned the spike Ɵmes of each neuron into 10 ms bins (10 bins for feeding; 50 bins for drinking). Neurons with a mean firing rate < 1.0 spike/s were excluded and the resultant vectors were smoothed using a Gaussian kernel with a 10 ms standard devia Ɵon (SD). To determine the dimensionality of the latent variable, we used 3-fold cross-validaƟon to find the value of m which maximized the likelihood of the data. We then obtained an FA model by fiƫ ng an m-dimensional latent factor model. FA was performed using trial-averaged data to examine the direcƟon-relevant latent factor responses (i.e., neural populaƟon trajectories). We quanƟfied direcƟonal differences in populaƟon acƟvity by calculaƟng the Euclidean distance over m latent factors) between trial-averaged neural popula Ɵon trajectories for each unique direcƟon pair (drinking = 3 pairs; feeding = 28 pairs). This analysis was performed for every 10 ms bin throughout each trial. To assess differences between experimental condiƟons (control vs. nerve block) or corƟcal region (MIo vs. SIo), we applied two-sample t-test to the mean inter- trajectory distances across all direcƟon pairs. We further characterized the neural space spanned by populaƟon acƟvity by measuring the cumulaƟve Euclidean distance travelled by trajectories from start to end of a trial. To control for potenƟal sampling biases, we implemented two criƟcal validaƟon procedures. First, we addressed the varying trial counts across direc Ɵons in the feeding task by performing Factor Analysis with standardized samples (N = 80 trials per direc Ɵon) through random subsampling repeated 10 Ɵmes. We then compared the cumulaƟve explained variance between the full and subsampled datasets. Second, we controlled for populaƟon size differences by subsampling MIo and SIo neurons to equivalent counts, enabling unbiased comparison of Factor Analysis results between corƟcal regions. To compare behaviors, we performed FA on only caudal MIo neurons that remained stable across both feeding and drinking recording sessions (N = 20). These neurons were determined through a stability test (Dickey et al., 2009), which compared the average waveform and interspike interval for both datasets. We analyzed two subsets of kinema Ɵc data : data collected within 200 ms surrounding minimum gape, and data from trials where the 3D angle measured 100 ms a Ō er minimum tongue protrusion was between -5 and +5 degrees. Decoding tongue direc Ɵ on. The ability to predict tongue direc Ɵon from spiking ac Ɵvity of MIo and SIo neurons was evaluated using a K-nearest neighbor (KNN) classifier. The Euclidean distance was used to idenƟfy nearest neighbors, and the number of nearest neighbors used was K = 7. This K value was determined aŌ er tesƟng different Ks which yielded comparable results. The feature .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint was the firing rate of each neuron over each trial: every 100 ms throughout feeding sequence, or 100 ms centered at minimum tongue protrusion during drinking. As a more direct comparison to the drinking, feeding direcƟons were split into three groups represenƟng leŌ , middle, and right movement direcƟons. The decoder was trained on 80% of trials and tested on the remaining 20%, then decoder performance was determined by the percentage of test trials where the direc Ɵon of movement was correctly decoded from the neural data. We ran 100 iteraƟons of the classifier using a different set of randomly selected training and test trials then calculated the average performance. The same sets of training and test trials were used for decoding from simultaneously recorded MIo and SIo data. However, our recorded populaƟons were of variable sizes, and decoding performance was found to be related to the number of neurons in the ensemble. Because the smallest popula Ɵon of neurons we recorded was 28, we selected 28 random neurons from the larger populaƟons for each iteraƟon. Based on the posiƟve relaƟonship between populaƟon size and decoding accuracy, we expect that performance would increase with more neurons. T hese results will show whether tongue direc Ɵon can be decoded from a very small number of neurons. We fit a linear regression model with interacƟons to compare decoding performance across the other variables in the experiment. To determine if a mixed populaƟon of MIo and SIo neurons performs be Ʃ er than the pure popula Ɵons, we started with the full MIo populaƟon and systemaƟcally replaced 25 MIo neurons with an equal number of SIo neurons. We repeated this replacement over 100 iteraƟons, each with a different random selecƟon of neurons, and decoded tongue direc Ɵon using the KNN classifier. We compared the average decoding performance of these slightly different mixed populaƟons to the baseline of running the decoder with the full MIo populaƟon. We also decoded tongue direcƟon using a long short-term memory (LSTM) network (Hochreiter and Schmidhuber, 1997; Glaser et al., 2020; Laurence-Chasen et al., 2023; Hahn and Arce- McShane, 2024) implemented in MATLAB's Deep Learning Toolbox. For the feeding task, we analyzed 3D tongue movement direcƟon (relaƟve to the sagi Ʃ al plane) at 100 ms intervals. For the drinking task, we categorized tongue direcƟon every 500 ms as either leŌ (-45°), middle (0°), or right (45°). For each neural populaƟon, we created a 2D array containing spike counts for these Ɵme intervals (neurons × intervals). We randomly selected five groups of 28 neurons with replacement from each popula Ɵon. Using 5 -fold cross-validaƟon, we trained an LSTM network (400 hidden units, 50 training epochs) on 85% of the intervals. The network was then tested on the remaining 15% of neural data in a stepwise manner, producing a sequence of predicted tongue direcƟons. We used mean R² to measure predicƟon accuracy.

Previous research has invesƟgated direcƟonal tuning of OSMCx neurons during trained tongue- protrusion tasks. Our study extends this work by inves ƟgaƟng two natural, untrained behaviors: feeding and licking ("drinking") from a spout. These behaviors provide dis Ɵnct contexts for .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint studying tongue coordina Ɵon - feeding allows unrestricted tongue movement, while drinking constrains movement direcƟon based on spout locaƟon. This comparison enables us to examine how OSMCx coordinates tongue movements under different behavioral contexts. Neuronal modulaƟ on paƩ erns differ between MIo and SIo. Many neurons exhibited significant modulaƟon of spiking acƟvity to tongue direcƟon (bootstrap, p < 0.05), though there were diverse paƩ erns. Figure 2 shows peri-event Ɵme histograms (PETHs) of two example neurons related to tongue movements during feeding and drinking. In feeding, both neurons showed complex oscillatory firing, with notable peaks between -0.05s to 0.1s for upward movements to the right (MIo, Fig. 2A-leŌ ) and posterior movements to the leŌ (SIo, Fig. 2A-right). For drinking there was clear separaƟon between different direcƟons in the MIo neuron , with the leŌ (green) exhibiƟng the highest ac Ɵvity (Fig. 2B- leŌ ), while SIo showed oscillatory pa Ʃ erns with less dis Ɵnct separaƟon between spout locaƟons, but higher overall acƟvity (Fig. 2B-right). Like the arm region, the tuning curves of direc Ɵonally modulated MIo and SIo neurons fit the cosine tuning funcƟon (F-test, p < 0.05, feeding: MIo = 86%, SIo = 75%). Figure 3A maps a neuron’s firing rate for tongue movements in leŌ -right, inferior-superior, and posterior-anterior axes. Here, a MIo neuron is strongly tuned to posterior-anterior and inferior-superior direcƟons, while remaining unresponsive to leŌ -right movements during natural feeding. Many of the recorded neurons in each populaƟon behaved in a similar fashion, with peaks most frequently observed toward the anterior and superior direcƟons. This observaƟon was consistent with the tongue movements being most frequent in the Anterior Superior direcƟons, followed by the Posterior Inferior (Figure 3 – figure supplement 1. The varying neuronal responses to tongue direc Ɵon c ould not be aƩ ributed to variability in their firing , as the distribuƟon of the Fano factor was similar across direcƟons (Kruskal-Wallis, p > 0.1 for all except SIo control drinking p = 0.06). The proporƟon of direc Ɵonally tuned neurons was higher in the feeding vs. drinking task (Chi- square, p < 0.05, feeding: 72%, drinking: 66%) and differed significantly between MIo and SIo during the feeding task in both subjects (Chi-square, p < 0.001). In rostral and caudal MIo, 80% of neurons were modulated to 3D direcƟon (bootstrap, p < 0.05, Fig. 3B , leŌ ), compared to 52% in areas 1/2 and 3a/3b. Notably, fewer MIo neurons showed direcƟonal tuning during swallows compared to chewing, while SIo neurons maintained consistent proporƟons (Figure 3 – figure supplement 2; Chi-square, MIo: p 0.1). During drinking, the propor Ɵon of direcƟonally modulated neurons was more similar between regions (69% in MIo vs. 60% in SIo: Chi-square, p > 0.05, Fig. 3B right). Mean-matched Fano factor was significantly lower in MIo than SIo in both tasks (Wilcoxon rank sum test, p < 0.001). We considered that the difference in the direcƟonal tuning between the two behaviors could be due to the different Ɵme intervals used for each task since the period around minimum tongue protrusion in the drinking may contain more of the sensory inputs from the previous lick. However, when sampling spiking acƟvity from an earlier period in feeding , the percentage of direc Ɵonally tuned SIo neurons was s Ɵll .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint significantly lower than MIo (Chi-square, p 0.1). Further analysis of the tongue’s lateral (yaw) and ver Ɵcal (pitch) components during feeding revealed addiƟonal insights. Figure 4A shows peak acƟvity of a neuron in MIo and in SIo at varying degrees of pitch and yaw. Overall, more neurons responded to pitch than yaw (Fig. 4B), with MIo showing a higher proporƟon of neurons tuned to both components compared to SIo (Chi-square, yaw: p < 0.08, pitch: p < 0.001), consistent with our 3D direc Ɵon results. MIo neurons exhibited sharper and narrower tuning curves than the broader tuning curves observed in SIo (Fig. 4A, see Supplementary file 2). The peak of the PD distribuƟ on coincides with leŌ ward tongue movements. The distribuƟon of preferred direcƟons provides insight into how tongue muscles are coordinated during movement. Intrinsic and extrinsic tongue muscles are involved in shaping the tongue (e.g., elongaƟon, broadening) and posiƟoning the tongue (e.g., protrusion/retracƟon, eleva Ɵon/depression), respecƟvely. These muscles receive bilateral motor innervaƟon except for genioglossus. Straight tongue protrusion requires the balanced acƟon of the right and leŌ genioglossi while the lateral protrusion involves primarily the contralateral genioglossus. Given this unilateral innerva Ɵon paƩ ern, we hypothesized that le Ō MIo/SIo neurons would preferen Ɵally respond to le Ō ward tongue movements, corresponding to right genioglossus acƟvaƟon. During feeding, MIo and SIo showed non-uniform distribuƟon of preferred direc Ɵons across a unit sphere (Fig. 5; Rayleigh test, p 0.1, mean ± 1 SD: MIo: 0.533 ± 0.3, SIo: 0.604 ± 0.5). As hypothesized, m ost neuronal popula Ɵons showed peaks in PD distribuƟons toward leŌ ward tongue movements, except in Monkey R's SIo (Fig. 6A). Similar results were found with the distribuƟons of preferred yaw during feeding (Supplementary file 2, Fig. 4). While feeding showed comparable PD distribu Ɵons between MIo and SIo in both subjects (circular k-test, p > 0.1), drinking revealed significant differences between regions in Monkey R (Chi-square, p 0.09). Monkey Y maintained predominantly leŌ -directed PDs across both tasks, while Monkey R showed more balanced le Ō - right PDs during drinking, sugges Ɵng poten Ɵal involvement of addi Ɵonal muscles beyond the right genioglossus. Neural populaƟ on trajectories differed based on task and corƟ cal regions . We analyzed direcƟonal tuning at the popula Ɵon level using Factor Analysis (FA) on simultaneously recorded neurons to extract neural trajectories and idenƟfy paƩ erns in their shared acƟvity. In both tasks, neural trajectories of popula Ɵon ac Ɵvity in both MIo and SIo exhibited robust direc Ɵonal informaƟon; inter-trajectory distances of all unique direcƟon pairs were significantly higher than zero (Fig. 7, t-test, p < 0.001, for all comparisons in both subjects and region). Notably in the feeding task, MIo and SIo showed smaller inter-trajectory distances between Anterior-Posterior .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint paired trajectories during feeding (e.g., AntSupL-brown and PostSupL-green) compared to a greater separaƟon between other direcƟonal pairs (Fig. 7-top, t-test, p 0.1). CumulaƟve explained variance for the first three factors was higher in feeding (MIo: 82%, SIo: 81%) than in drinking (MIo: 74%, SIo: 63%) when all neurons were used for the factor analysis (Fig. 7). Similar results were obtained when equal number of neurons were used (Fig. 7- figure supplement 1). To control for factors such as different neurons and kinemaƟcs that might influence the results, we performed factor analysis on stable neurons across both tasks using all trials (Fig. 7- figure supplement 2A) and using trials with similar kinemaƟcs (Fig. 7- figure supplement 2B). While the general shape of the populaƟon trajectory was preserved across tasks, the inter-trajectory distance between them was significant (t-test, p 0.05) and longer tongue displacement in drinking ( p < 0.001, mean ± 1 SD: feeding: 27.5 ± 9.8, drinking: 47.6 ± 11). When trials were limited to those with similar direc Ɵonal angles (±5 degrees), the difference in trajectory length was no longer observed but the inter-trajectory distance between tasks remained significantly different (t-test, p < 0.001, mean ± 1 SD: 0.6967 ± 0.0793). MIo populaƟon trajectories followed a circular path and exhibited consistent paƩ erns based on direcƟonal components. For example (Fig. 7 top), trajectories with upward (Sup) components (AntSupL/R, PostSupL/R) rotated opposite from trajectories with downward (Inf) components (AntInfL/R, PostInfL/R). In feeding, Factors 1 and 2 captured superior-inferior and right- leŌ direcƟons, respec Ɵvely. In drinking, MIo trajectories exhibited consistent rotaƟonal direc Ɵon regardless of spout locaƟon (Fig. 7 boƩ om leŌ ), while exhibiƟng disƟnct separaƟon of trajectories for leŌ , center, and right spout-directed tongue movements clustering at approximately -0.5, 0, and 0.5 posiƟons along the Factor 2 axis, respecƟvely. Indeed, inter-trajectory distances in Factor 1 were significantly higher in feeding (t-test, p < 0.001, mean ± 1 SD: 0.4628 ± 0.0246) than in drinking (mean ± 1 SD: 0.1286 ± 0.0610 ), indica Ɵng that Factor 1 resembled direc Ɵonal informaƟon in feeding but a condiƟon-independent feature of populaƟon acƟvity in drinking. The latent factors revealed a clear organizaƟonal principle: Factor 1 predominantly captured superior- inferior direc Ɵonal components in feeding , while Factor 2 primarily represented le Ō -right direcƟonal components of tongue movement in both tasks. Similar to previous findings (Russo et al., 2018), SIo trajectories in both feeding and drinking showed stark differences from MIo as they were more tangled and exhibited less direct (i.e., sharp turns) paths (Fig. 7 right). Unlike MIo trajectories, SIo trajectories spanned a smaller neural space, had variable distances between trajectories, and showed inconsistent paƩ erns based on direcƟonal components . Q uanƟtaƟve analysis revealed greater separa Ɵon between trial- averaged populaƟon trajectories in MIo compared to SIo (Fig. 7 inset, t-test, p < 0.01, mean ± 1 SD: MIo: 0.34 ± 0.03; SIo: 0.12 ± 0.02). These results were consistent with significantly longer distance travelled by MIo populaƟon trajectories compared to SIo in both tasks for Monkey R only .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint (t-test, p < 0.001, mean ± 1 SD: feeding: MIo: 0.43 ± 0.07; SIo: 0.26 ± 0.09; drinking: MIo: 2.85 ± 0.42; SIo: 1.36 ± 0.52). T he regional differences cannot be a Ʃ ributed to fewer SIo neurons used (Figure 7 – figure supplement 3). Effects of nerve block SensaƟon plays a key role in tongue posi Ɵoning and movements for natural behaviors. During ingesƟon, tacƟle feedback is necessary for loca Ɵng the bolus, preven Ɵng tongue bites, feeling where the drinking spout is, and idenƟfying when it is safe to swallow. To evaluate the role of oral sensaƟon, we used a bilateral oral nerve block to temporarily eliminate tacƟle sensaƟon in the oral cavity and observe how the control of tongue movement was impacted. Below, we show how the loss of sensa Ɵon affected both tongue kinemaƟcs and direc Ɵonal tuning of neurons during feeding and drinking. To verify that differenc es between the control and nerve block condi Ɵons were due to the loss of sensory feedback and not because of other factors such as sedaƟon and injecƟon, a sham experiment was conducted where saline was administered to the injecƟon sites instead of nerve block. No significant changes to tongue kinemaƟcs were observed following the sham experiments (Fig. 8). Changes to t ongue kinema Ɵ cs. In feeding, the mean and overall spread of direc Ɵons w ere significantly different between the control and nerve block condiƟons (t-test, p < 0.01 and f-test, p < 0.001). There was a shiŌ towards a smaller range of 3D direcƟons in Monkey R, whereas there was a shiŌ towards a broader distribuƟon in Monkey Y under the nerve block condiƟon (Fig. 8A). The posiƟons of maximum protrusion of the tongue during drinking, i.e., the endpoints, were also affected by the loss of sensa Ɵon. These endpoints represent the planned target posi Ɵon of the tongue to receive the juice reward from a specific spout. In the control drinking task, the endpoints for each spout loca Ɵon were very dis Ɵnct. In contrast, the endpoints of tongue movements in nerve block exhibited a greater overlap across loca Ɵons and more variance in all three axes of moƟon, i.e., Posterior-Anterior, Inferior-Superior, and LeŌ -Right (Fig. 8B). Compared to the control, the trajectories of the tongue Ɵp in the nerve block condi Ɵon during drinking had a smaller range of Le Ō -Right values. Visually, the tongue trajectories toward the different spout loca Ɵons were messier and less dis Ɵnct in failed cycles where the tongue Ɵp missed the loca Ɵon of the correct spout by more than 2 SD from the mean (Fig. 9). In both monkeys there was a significant increase in the average distance from the mean endpoint posiƟon, though this difference was much greater in Monkey R ( Fig. 8C). We noted a difference between subjects in the frequency of failed cycles and the range of leŌ -right tongue movements under nerve block. This may reflect a possible compensatory strategy of reaching the drinking spouts with an adjacent region of the tongue, instead of contacƟng the right or leŌ spout with the ipsilateral tongue in Monkey R. We observed a decrease in the average speed (t-test, Monkey R: p 0.1) and an increase in the variance of the speed (F-test p < 0.001) of tongue movement during drinking with nerve block. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Decreased direcƟ onal tuning of MIo and SIo neurons . Loss of oral sensa Ɵon also affected the proporƟon of direcƟonally tuned neurons and the overall distribuƟon of PDs, though the paƩ ern of changes differed between subjects. Following nerve block, MIo and SIo showed significant decreases in the proporƟon of direcƟonally modulated neurons across both tasks (Fig. 10A; Chi- square, MIo: p < 0.001, SIo: p < 0.05). To confirm this effect was not merely due to altered kinemaƟcs, we conducted parallel analyses using carefully subsampled trials with matched kinemaƟc profiles from both control and nerve -blocked condi Ɵons. This controlled analysis confirmed the persist ent decrease in direc Ɵonal tuning during nerve block (Figure 10 – figure supplement 1). We further invesƟgated whether neurons gaining or losing direc Ɵonal selecƟvity differed across regions. During feeding, MIo and SIo exhibited similar propor Ɵons of neurons gaining or losing direcƟonal tuning (Fig. 10B-top row, Chi-square, p > 0.1). The drinking task revealed subject- specific differences (Fig. 10B- boƩ om row) : in Monkey R, significantly more neurons lost direcƟonal tuning in SIo compared to MIo (p < 0.01), while in Monkey Y , SIo showed a higher proporƟon of neurons gaining direcƟonal tuning than MIo (p 0.1), whereas Monkey Y showed significantly higher percentages of neurons gaining direc Ɵonality during drinking than feeding in both MIo and SIo ( p < 0.05). InteresƟngly, there was a large proporƟon (40%) of SIo neurons in Monkey Y that gained direcƟonal tuning following sensory loss compared to Monkey R (8%) during drinking. Nerve block significantly altered PD distribuƟons during both tasks. During feeding, MIo neurons in both subjects exhibited a significant clockwise shiŌ in mean PD toward the center (0°), resulƟng in more uniform distribuƟons (Fig. 11A; circular k-test, p < 0.01). In contrast, SIo neurons showed subject-specific responses, with only Monkey R demonstraƟng a significant counterclockwise shiŌ (p < 0.05). During drinking under nerve block, MIo neurons displayed subject-dependent direcƟonal shi Ō s. In Monkey R, the proporƟon of neurons with rightward PDs decreased and increased in all other direcƟons whereas Monkey Y showed the opposite with increased neurons with rightward PDs (Fig. 11B; Chi-square, Y: p = 0.04). Meanwhile, SIo neurons consistently shiŌ ed rightward in both animals (Chi-square, R: p = 0.02), sugges Ɵng differenƟal regional responses to peripheral deafferentaƟon. SeparaƟ on of neural popula Ɵ on trajectories was reduced in MIo. DisrupƟon of tac Ɵle inputs during feeding had opposite effects on MIo and SIo. Neural populaƟon trajectories in MIo showed reduced inter-trajectory distances during nerve block compared to control condi Ɵons (Fig. 12, leŌ , t-test, p 89% of pairs), whereas SIo exhibited increased inter-trajectory distances (Fig. 12 top right, t-test, p 75% of pairs). In drinking, inter-trajectory distances in both MIo and SIo were significantly reduced across all pairs (two-tailed t-test, p 0.1). .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint The effects of nerve block on the distance travelled by populaƟon trajectories were inconsistent across subjects and behavior. Following nerve block, the total distance travelled by SIo trajectories became longer in different behaviors for a specific subject (t-test, Monkey R feeding: p < 0.001; Monkey Y drinking: p < 0.05). In contrast, MIo trajectories became shorter in drinking (Monkey Y , p < 0.01). PopulaƟ on decoding of tongue direc Ɵ on. To assess direc Ɵonal informaƟon within popula Ɵon acƟvity further, we implemented two decoding approaches to predict tongue movement direcƟon from neuronal spiking paƩ erns: k-nearest neighbor (KNN) classifier and long short-term memory (LSTM) neural network. Consistent with previous study on a cued tongue protrusion task (Arce et al., 2013), we found that the 3D direcƟon of tongue movements in naturalisƟc behaviors could be decoded from simultaneously recorded MIo and SIo popula Ɵons. The KNN classifier successfully decoded 3D tongue movement direc Ɵon above chance level across behaviors and experimental condiƟons (Fig. 13A). Results of analyses using m ulƟple linear regression model with interacƟons revealed several key factors affecƟng decoder performance: behavior type (p < 0.001, drinking outperformed feeding by 11 %), corƟcal region ( p < 0.001, MIo exceeded SIo by 13%), and inter-subject variability with behavior ( p < 0.001, 12% higher for Monkey R during drinking, comparable accuracy during feeding). Notably, disrupƟng tac Ɵle sensa Ɵon through nerve block did not significantly impair KNN classifier performance ( p > 0.1). The results using LSTM were different from that of KNN; decoding tongue direc Ɵon using LSTM showed substanƟally higher performance when using MIo neural ac Ɵvity (mean R2 ± 1 SD: control: 0.46- 0.81 ± 0.1, nerve block: 0.26-0.7 ± 0.1) compared to SIo (mean R2 ± 1 SD: control: 0.12-0.43 ± 0.1, nerve block: 0.05-0.17 ± 0.07) across all experimental condiƟons (Fig. 13B, t-test, p < 0.001). This regional difference became par Ɵcularly pronounced during nerve block, where SIo decoding accuracy decreased substan Ɵally more than MIo, sugges Ɵng differen Ɵal reliance on tac Ɵle feedback between these cor Ɵcal regions. Combining MIo and SIo showed significantly be Ʃ er decoder performance compared to performance using neuronal populaƟons separately (mean R2 ± 1 SD: control: 0.78-0.92 ± 0.05, nerve block: 0.58-0.91 ± 0.05, p < 0.001) in feeding, but not drinking. To address the potenƟal confound of varying populaƟon sizes between MIo and SIo, we standardized comparisons by downsampling all popula Ɵons to match our smallest recorded group (N = 28 neurons). Decoding accuracy improved by up to 10% when using all neurons in MIo or SIo compared to using subsamples of neurons. Decoding using LSTM showed consistently higher accuracies in feeding compared to drinking regardless of the length of intervals used (100 ms, 500 ms), behavioral window (  500 ms relaƟve to minimum protrusion, between maximal protrusions), and direcƟonal angles (actual vs. {-45, 0, 45}). These results suggest that conƟnuous and non-linear decoding is beƩ er suited for feeding than drinking behavior. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint

Our study invesƟgated the representaƟon of 3D direcƟonal informaƟon in MIo and SIo during natural feeding and drinking behaviors, and how corƟcal representaƟons change when tac Ɵle sensaƟon is disrupted . By simultaneously recording large-scale corƟcal acƟvity and 3D tongue kinemaƟcs, we revealed a nuanced neural encoding of tongue movement direc Ɵon that varies systemaƟcally across corƟcal regions, behaviors, and sensory feedback condiƟons. We found that a substanƟal propor Ɵon of neurons exhibit direcƟonal tuning characterized by diverse tuning curve proper Ɵes (PD, tuning shape, modulaƟon depth ). N eural popula Ɵon trajectories demonstrated disƟnct paƩ erns across different movement direcƟons. DirecƟonal tuning in both individual- and popula Ɵon-level is more robust in MIo. Following sensory loss, alteraƟons in tongue kinema Ɵcs were accompanied by changes in direcƟonal informa Ɵon in MIo and SIo, manifesƟng as modifica Ɵons in both individual neuron tuning characteris Ɵcs and the broader dynamics of populaƟon-level neural acƟvity. Differences across behaviors. In the present study, results from the more natural drinking are consistent with previous findings that MIo and SIo encode tongue direcƟon during a trained protrusion task (Murray and Sessle, 1992; Sessle et al., 2005a; Arce et al., 2013). Our study extends this knowledge by invesƟgaƟng the dynamics of direcƟonal tuning of individual and populaƟon of neurons in OSMCx to 3D tongue direc Ɵon during naturalisƟc behaviors. Unlike previous similar studies, the monkeys were not trained to reach specific targets and were instead allowed to eat and drink rela Ɵvely naturally. By comparing two natural isƟc behaviors, we found that the direcƟonal informaƟon in OSMcx was higher in feeding than in the drinking task, as seen in higher proporƟon of direc Ɵonally tuned neurons, cumulaƟve variance explained by latent factors, and decoding accuracies. That direcƟonal tuning is modifiable is consistent with previous findings in primate motor cortex where direc Ɵonal tuning was modulated by movement parameters such as speed, posture, distance (Aflalo and Graziano, 2006) and by varying task contexts such as availability of prior informa Ɵon (Rickert et al., 2009), individual vs. segmented arm movements (Ben-Shaul et al., 2004) , one- vs two-target reaching (Ebina et al., 2024). A high degree of similarity in neural modes have been reported across different wrist tasks in 1-D and 2D (Gallego et al., 2018). This suggests that in our study, feeding and drinking may reflect more disƟnct biomechanical constraints and sensor imotor requirements compared to the wrist tasks. In feeding, the tongue moves in varied direc Ɵons to posi Ɵon the food between le Ō -right toothrows during chewing, whereas in the drinking task, the tongue moves to discrete and fixed spout locaƟons. AddiƟonally, feeding and drinking engage the jaw differently. During feeding, the jaw moves more extensively and in mulƟple direcƟons, while jaw movements during drinking are smaller and primarily verƟcal. Lastly, the tongue- jaw coordina Ɵon differs between tasks; maximum tongue protrusion coincides with maximum gape in drinking but with minimum gape in the feeding behavior (Laurence-Chasen et al., 2022; Punacha et al., 2024). Indeed, MIo .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint trajectories’ predominant latent factor contained direc Ɵonal informa Ɵon in the feeding task, resembling superior- inferior direc Ɵonal components which rotated in opposite direc Ɵons. In contrast, MIo trajectories in drinking exhibited a consistent rotaƟonal direc Ɵon regardless of spout loca Ɵon (Fig. 7). This may reflect a predominant non- direcƟonal informa Ɵon such as condiƟon-independent Ɵme-varying spiking acƟvity during drinking (Kaufman et al., 2013; Kobak et al., 2016; Arce-McShane et al., 2023). Comparison between MIo and SIo. DirecƟonal tuning of neurons during feeding showed a notable disparity between MIo and SIo, suggesƟng that MIo carries more robust direcƟonal informaƟon for tongue movements in feeding tasks. The similar propor Ɵons of direc Ɵonally tuned MIo and SIo neurons in the drinking task studied here were consistent with previous findings (Arce et al., 2013). At the populaƟon level, MIo trajectories showed more consistent rotaƟonal paƩ erns and greater inter-trajectory separaƟon than those in SIo. Consistent with results from previous studies (Michaels et al., 2016; Seely et al., 2016; Russo et al., 2018), MIo trajectories exhibited low tanging and smoother dynamics than SIo trajectories. These may suggest that the low tangling in MIo confers noise robustness while the higher tangling in SIo reflects variability in the tacƟle signals received by SIo during feeding. Consistent with our previous study (Laurence-Chasen et al., 2023), decoding from MIo yielded higher accuracies than from SIo in both behaviors. These results support the well-established role of MIo in the control of movement parameters, especially direcƟon. Varying tongue shape, tongue’s contact with varying bolus properƟes (size and texture) and other oral structures (palate, teeth) may weaken the direcƟonal signal contained in SIo acƟvity. Thus, small differences in tongue kinemaƟcs might create large differences in sensory signals across trials. When looking at trial-averaged signals, this natural variability could make the neural response paƩ erns appear less precise or specific than they are. These are consistent with our findings that for both tasks, spiking variability was higher in SIo, and that the variance accounted for was lower in SIo populaƟon acƟvity compared to MIo. Laterality in OSMCx. Similar to previous results in arm motor cortex (Lillicrap and ScoƩ , 2013), we observed non- uniform PD distribu Ɵons consistent with the frequency distribu Ɵon of tongue movements, suggesƟng that neural populaƟons contain informaƟon that reflects the anatomical constraints of the tongue . The highest frequency of both observed direc Ɵons and direc Ɵonal tuning peaks were in the anterior and superior direcƟons. We addiƟonally found that the peak of the PD distribuƟon, especially in feeding, coincides with leŌ ward tongue movements, suggesƟng the presence of laterality in the PDs of OSMCx neurons. Previous results in humans examined using fMRI reported that hemispheric differences in sensorimotor ac Ɵvity during voluntary tongue movements are related to the preferred chew side (Shinagawa et al., 2003). This was not the case in our study as the preferred chew side was the same for both monkeys (i.e., right side), despite differences in the predominant PD. It is possible that the difference between the two subjects is related to the difference in recording locaƟons, with Monkey Y’s being more lateral .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint and therefore closer to the swallow area of the cortex than Monkey R’s (Supplementary file 1, Fig. 1). Monkey Y had a higher propor Ɵon of neurons that were tuned to tongue direc Ɵon during feeding compared to Monkey R (Figure 3 – figure supplement 2), but fewer during drinking. An avenue for further study could be a unilateral nerve block on the preferred side to measure how the unaffected side of the tongue compensates for the lack of sensa Ɵon in the affected side. A previous study found that unilateral lingual nerve transec Ɵon in pigs alters the coordina Ɵon of the ipsilateral tongue side during chewing (Montuelle et al., 2020). The tongue is a complex group of muscles, with intrinsic muscles primarily contribuƟng to the shape of the tongue and extrinsic muscles contribu Ɵng more to the posi Ɵoning of the tongue. Therefore, it is possible that the neurons which are strongly tun ed to tongue direc Ɵon have direct connec Ɵons to the extrinsic muscles on the ipsilateral side. Looking at how each side of the tongue responds independently to unilateral nerve block, and how this interacts with direc Ɵonal preference may give us more informaƟon about how the unique structure of the tongue is coordinated. Role of tacƟle feedback. Previously, we reported that the administraƟon of bilateral nerve block to the sensory branches of the trigeminal nerve impaired feeding performance and tongue jaw coordinaƟon (Laurence-Chasen et al., 2022). The present study extends these findings by showing that direcƟonal movement of the tongue ( kinemaƟcs) and the direcƟonal informaƟon in both MIo and SIo are also affected by the loss of tacƟle inputs from the tongue and other structures of the oral cavity (e.g., palate, teeth, gingiva). These findings highlight the cri Ɵcal role of sensory informaƟon in sensorimotor control in general (Dadarlat et al., 2015; Delhaye et al., 2018) and in the representaƟon and computaƟon of direcƟonal signals for controlling tongue movements. MIo and SIo neurons, which respond to t acƟle and propriocepƟve inputs from the tongue (Huang et al., 1989; Lin et al., 1994b; Toda and Taoka, 2002, 2004, 2006; Arce et al., 2013; Cerkevich et al., 2014), use sensory informaƟon to plan and adjusts tongue movements to achieve contact with the spout or posiƟon the bolus appropriately at different stages of the feeding sequence. Without tacƟle feedback, subjects may rely on alternaƟve sensory cues like taste for locaƟng the spout or bolus (Todrank and Bartoshuk, 1991) . In a recent optogeneƟc inhibiƟon study on licking in mice, it was found that the tongue/jaw regions of the somatosensory cortex were necessary for proper tongue targeƟng but not for the core motor capabili Ɵes of the tongue (Xu et al., 2022). The reduced range of tongue mo Ɵon we observed likely stems from sensory loss rather than motor impairment. While our experimental setup did not eliminate visual feedback that monkeys might use to readjust tongue posiƟon in the drinking task, oral sensory loss alone had a significant effect on the monkeys’ performance in feeding, as the tongue was out of sight within the oral cavity. Individual differences were notable in our study possibly due to differences in the electrode array placement or compensatory strategies. This individual variability suggests that studying addiƟonal subjects would provide valuable insights into how OSMCx adapts following sensory disrupƟon. In our study, head posi Ɵon and hand movement were restrained, to eliminate .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint contribuƟons of hand-to-mouth movements when handling food or drink. The hand and orofacial corƟcal areas are anatomically adjacent and highly interconnected (Forrester and Rodriguez, 2015), and researchers have found a neural region in mice that coordinates hand-to-mouth movements during natural feeding (An et al., 2022). Truly natural feeding would involve holding food up to the mouth, as well as free head movement, which would make tracking of the marker posiƟons difficult under this experimental setup. Advances in tracking tongue movements would be necessary to study more complex feeding sequences. Clinical implicaƟ ons This study offers new informa Ɵon about the important role of sensorimotor integra Ɵon in controlling tongue direc Ɵon during natural behaviors. There is a high degree of direc Ɵonal informaƟon contained in the spiking acƟvity of the orofacial cortex, especially in the motor areas. The effect of the bilateral nerve block serves to enhance our understanding of the processes affected by oral sensorimotor dysfuncƟons such as trigeminal neuropathies. It demonstrates the importance of oral sensa Ɵon for supporƟng the full range of direc Ɵonal moƟon but also shows that significant direcƟonal informaƟon can be extracted even in the absence of tac Ɵle feedback. This type of knowledge can inform the diagnosis and rehabilita Ɵon of orolingual dysfunc Ɵons, following stroke or glossectomy. There have also been advancements in brain-computer interface (BCI) by decoding the real- Ɵme signals of arm region of the motor cortex to control prosthe Ɵc arm movement (Collinger et al., 2013; Vilela and Hochberg, 2020) or muscle sƟmulaƟon (Ethier and Miller, 2015), as well as efforts to restore sensory feedback by s ƟmulaƟng correct areas of somatosensory cortex in response to sensors on a prosthe Ɵc (Tabot et al., 2013; Flesher et al., 2021). That the OSMCx, par Ɵcularly MIo, can rapidly decode tongue direc Ɵon during natural behaviors is significant for developing neuroprostheƟc control or soŌ prostheƟcs. Ethics statement Experiments were performed in the University of Chicago XROMM Facility. All protocols were approved by the University of Chicago Animal Care and Use Commi Ʃ ee and complied with the NaƟonal InsƟtutes of Health Guide for the Care and Use of Laboratory Animals. Funding informaƟ on This research was supported by NaƟonal InsƟtutes of Health grants from the NaƟonal InsƟtute on Aging, Grant Number: R01AG069227 (to F.I.A- M, PI) and the Na Ɵonal InsƟtute of Dental and Craniofacial Research, Grant Number: R01DE027236 (to F.I.A-M, PI). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NaƟonal InsƟtutes of Health. The funders had no role in study design, data collec Ɵon and interpreta Ɵon, or the decision to submit the work for publicaƟon. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Author contribuƟ ons Victoria B. Hosack: data analysis, wri Ɵng – original dra Ō , wri Ɵng – review and edi Ɵng, visualizaƟon. Fritzie I. Arce- McShane: conceptualiza Ɵon, methodology, inves ƟgaƟon, wri Ɵng – review and ediƟng, supervision, project administraƟon, funding acquisiƟon.

We thank J.D. Laurence- Chasen for data collec Ɵon and so Ō ware, ChrisƟna Hahn for assistance with LSTM analysis, as well as all members of the Arce-McShane Lab past and present, including Rebecca Junod, Hernando Fereira, Derrick Tang, Emma Lesser, Jared Luckas, Tricia Nicholson, Eric Hosack, for assistance with data collecƟon and processing.

Aflalo TN, Graziano MS (2006) ParƟal tuning of motor cortex neurons to final posture in a free-moving paradigm. Proc Natl Acad Sci USA 103:2909–2914 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/16473936. Ajemian R, Bullock D, Grossberg S (2000) KinemaƟc Coordinates In Which Motor CorƟcal Cells Encode Movement DirecƟon. J Neurophysiol 84:2191–2203 Available at: hƩ ps://doi.org/10.1152/jn.2000.84.5.2191. An X, Matho K, Li Y , Mohan H, Xu XH, Whishaw IQ, Kepecs A, Huang ZJ (2022) A corƟcal circuit for orchestraƟng oromanual food manipulaƟon. bioRxiv. Arce F, Novick I, Mandelblat-Cerf Y , Israel Z, Ghez C, Vaadia E (2010) Combined adapƟveness of specific motor corƟcal ensembles underlies learning. Journal of Neuroscience 30. Arce FI, Lee J-C, Ross CF, Sessle BJ, Hatsopoulos NG (2013) DirecƟonal informaƟon from neuronal ensembles in the primate orofacial sensorimotor cortex. J Neurophysiol 110:1357–1369 Available at: hƩ p://dx.doi.org/10.1152/jn.00144.2013. Arce-McShane FI, Sessle BJ, Ram Y, Ross CF, Hatsopoulos NG (2023) MulƟple regions of primate orofacial sensorimotor cortex encode bite force and gape. Front Syst Neurosci. Avivi-Arber L, Sessle BJ (2017) Jaw sensorimotor control in healthy adults and effects of ageing. J Oral Rehabil 45:50–80 Available at: hƩ p://dx.doi.org/10.1111/joor.12554. Bach-y-Rita P , Kaczmarek KA, Tyler ME, Garcia-Lara J (1998) Form percepƟon with a 49-point electrotacƟle sƟmulus array on the tongue: a technical note. J Rehabil Res Dev 35:427–430 Available at: hƩ ps://www.ncbi.nlm.nih.gov/pubmed/10220221. Ben-Shaul Y , Drori R, Asher I, Stark E, Nadasdy Z, Abeles M (2004) Neuronal AcƟvity in Motor CorƟcal Areas Reflects the SequenƟal Context of Movement. J Neurophysiol 91:1748–1762 Available at: hƩ ps://doi.org/10.1152/jn.00957.2003. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Berens P (2009) CircStat : A MATLAB Toolbox for Circular StaƟsƟcs. J Stat SoŌ w 31. Brainerd EL, Baier DB, Gatesy SM, Hedrick TL, Metzger KA, Gilbert SL, Crisco JJ (2010) X-ray reconstrucƟon of moving morphology (XROMM): precision, accuracy and applicaƟons in comparaƟve biomechanics research. J Exp Zool A Ecol Genet Physiol 313A:262–279 Available at: hƩ p://dx.doi.org/10.1002/jez.589. Cerkevich CM, Qi H-X, Kaas JH (2014) CorƟcocorƟcal projecƟons to representaƟons of the teeth, tongue, and face in somatosensory area 3b of macaque monkeys. J Comp Neurol 522:546–572. Churchland MM et al. (2010) SƟmulus onset quenches neural variability: a widespread corƟcal phenomenon. NatNeurosci 13:369–378 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/20173745. Churchland MM, Cunningham JP , Kaufman MT, Foster JD, Nuyujukian P , Ryu SI, Shenoy K V (2012) Neural populaƟon dynamics during reaching. Nature 487:51–56 Available at: hƩ ps://doi.org/10.1038/nature11129. Collinger JL, Wodlinger B, Downey JE, Wang W , Tyler-Kabara EC, Weber DJ, McMorland AJC, Velliste M, Boninger ML, Schwartz AB (2013) High-performance neuroprostheƟc control by an individual with tetraplegia. Lancet 381:557–564 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/23253623. Cowley BR, Kaufman MT, Butler ZS, Churchland MM, Ryu SI, Shenoy K V, Yu BM (2013) DataHigh: graphical user interface for visualizing and interacƟng with high-dimensional neural acƟvity. J Neural Eng 10:066012 Available at: hƩ ps://dx.doi.org/10.1088/1741-2560/10/6/066012. Dadarlat MC, O’Doherty JE, Sabes PN (2015) A learning-based approach to arƟficial sensory feedback leads to opƟmal integraƟon. Nat Neurosci 18:138–144 Available at: hƩ ps://doi.org/10.1038/nn.3883. Delhaye BP , Long KH, Bensmaia SJ (2018) Neural Basis of Touch and PropriocepƟon in Primate Cortex. In: Comprehensive Physiology, pp 1575–1602 Available at: hƩ ps://doi.org/10.1002/cphy.c170033. Dickey AS, Suminski A, Amit Y , Hatsopoulos NG (2009) Single-unit stability using chronically implanted mulƟelectrode arrays. JNeurophysiol 102:1331–1339 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/19535480. Ebina T, Sasagawa A, Hong D, Setsuie R, Obara K, Masamizu Y , Kondo M, Terada S-I, Ozawa K, Uemura M, Takaji M, Watakabe A, Kobayashi K, Ohki K, Yamamori T, Murayama M, Matsuzaki M (2024) Dynamics of direcƟonal motor tuning in the primate premotor and primary motor corƟces during sensorimotor learning. Nat Commun 15:7127 Available at: hƩ ps://doi.org/10.1038/s41467-024- 51425-3. Ethier C, Miller LE (2015) Brain-controlled muscle sƟmulaƟon for the restoraƟon of motor funcƟon. Neurobiol Dis 83:180–190 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/25447224. Feilich K, Orsbon C, Gidmark N, Ross C (2021) Twist and chew: three-dimensional tongue kinemaƟcs during chewing in macaque primates. Biol LeƩ 17. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Flesher SN, Downey JE, Weiss JM, Hughes CL, Herrera AJ, Tyler-Kabara EC, Boninger ML, Collinger JL, Gaunt RA (2021) A brain-computer interface that evokes tacƟle sensaƟons improves roboƟc arm control. Science 372:831–836 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/34016775. Forrester GS, Rodriguez A (2015) Slip of the tongue: ImplicaƟons for evoluƟon and language development. CogniƟon 141:103–111 Available at: hƩ p://dx.doi.org/10.1016/j.cogniƟon.2015.04.012. Gallego JA, Perich MG, Naufel SN, Ethier C, Solla SA, Miller LE (2018) CorƟcal populaƟon acƟvity within a preserved neural manifold underlies mulƟple motor behaviors. Nat Commun 9:4233 Available at: hƩ p://dx.doi.org/10.1038/s41467-018-06560-z. Georgopoulos AP , KeƩ ner RE, Schwartz AB (1988) Primate Motor Cortex and Free Arm Movements to Visual Targets in Three-Dimensional Space. II. Coding of the DirecƟon of Movement by a Neuronal PopulaƟon. JNeurosci 8:2928–2937. Georgopoulos AP , Merchant H, Naselaris T, Amirikian B (2007) Mapping of the preferred direcƟon in the motor cortex. Proc Natl Acad Sci U S A 104:11068–11072 Available at: hƩ p://dx.doi.org/10.1073/pnas.0611597104. Glaser JI, Benjamin AS, Chowdhury RH, Perich MG, Miller LE, Kording KP (2020) Machine Learning for Neural Decoding. eNeuro 7:ENEURO.0506-19.2020 Available at: hƩ p://www.eneuro.org/content/7/4/ENEURO.0506-19.2020.abstract. Hahn C, Arce-McShane FI (2024) Impact of nerve block on the corƟcal decoding of tongue movements across axes of moƟon and marker regions. In: Program No. LBA005.27. 2024 Neuroscience MeeƟng Planner. Chicago, IL: Society for Neuroscience. Hiiemae KM, Palmer JB (2003) Tongue Movements in Feeding and Speech. CriƟcal Reviews in Oral Biology & Medicine 14:413–429 Available at: hƩ ps://doi.org/10.1177/154411130301400604. Hochreiter S, Schmidhuber J (1997) Long Short-Term Memory. Neural Comput 9:1735–1780. Horrocks EAB, Rodrigues FR, Saleem AB (2024) Flexible neural populaƟon dynamics govern the speed and stability of sensory encoding in mouse visual cortex. Nat Commun 15:6415 Available at: hƩ ps://doi.org/10.1038/s41467-024-50563-y. Huang CS, Hiraba H, Sessle BJ (1989) Input-output relaƟonships of the primary face motor cortex in the monkey (Macaca fascicularis). J Neurophysiol 61:350–362 Available at: hƩ ps://doi.org/10.1152/jn.1989.61.2.350. Kaufman MT, Churchland MM, Shenoy K V (2013) The roles of monkey M1 neuron classes in movement preparaƟon and execuƟon. J Neurophysiol 110:817–825 Available at: hƩ ps://doi.org/10.1152/jn.00892.2011. Knörlein BJ, Baier DB, Gatesy SM, Laurence-Chasen JD, Brainerd EL (2016) ValidaƟon of XMALab soŌ ware for marker-based XROMM. Journal of Experimental Biology. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Kobak D, Brendel W, ConstanƟnidis C, Feierstein CE, Kepecs A, Mainen ZF, Qi X-L, Romo R, Uchida N, Machens CK (2016) Demixed principal component analysis of neural populaƟon data. Elife 5:1–36 Available at: hƩ ps://elifesciences.org/arƟcles/10989. Lamm NC, De Felice A, Cargan A (2005) Effect of tacƟle sƟmulaƟon on lingual motor funcƟon in pediatric lingual dysphagia. Dysphagia 20:311–324 Available at: hƩ p://dx.doi.org/10.1007/s00455-005-0060-7. Laurence-Chasen JD, Arce-McShane FI, Hatsopoulos NG, Ross CF (2022) Loss of oral sensaƟon impairs feeding performance and consistency of tongue-jaw coordinaƟon. J Oral Rehabil 49:806–816 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/35514258. Laurence-Chasen JD, Manafzadeh AR, Hatsopoulos NG, Ross CF, Arce-McShane FI (2020) IntegraƟng XMALab and DeepLabCut for high-throughput XROMM. J Exp Biol 223:jeb226720 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/32665442. Laurence-Chasen JD, Ross CF , Arce-McShane FI, Hatsopoulos NG (2023) Robust corƟcal encoding of 3D tongue shape during feeding in macaques. Nat Commun 14:2991 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/37225708. Lillicrap TP , ScoƩ SH (2013) Preference DistribuƟons of Primary Motor Cortex Neurons Reflect Control SoluƟons OpƟmized for Limb Biomechanics. Neuron 77:168–179 Available at: hƩ ps://doi.org/10.1016/j.neuron.2012.10.041. Lin LD, Murray GM, Sessle BJ (1994a) FuncƟonal properƟes of single neurons in the primate face primary somatosensory cortex. I. RelaƟons with trained orofacial motor behaviors. J Neurophysiol 71:2377– 2390 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/7931522. Lin LD, Murray GM, Sessle BJ (1994b) FuncƟonal properƟes of single neurons in the primate face primary somatosensory cortex. I. RelaƟons with trained orofacial motor behaviors. J Neurophysiol 71:2377– 2390 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/7931522. Lozano CA, Kaczmarek KA, Santello M (2009) ElectrotacƟle sƟmulaƟon on the tongue: Intensity percepƟon, discriminaƟon, and cross-modality esƟmaƟon. Somatosens Mot Res 26:50–63 Available at: hƩ p://dx.doi.org/10.1080/08990220903158797. Mathis A, Mamidanna P , Cury KM, Abe T, Murthy VN, Mathis MW, Bethge M (2018) DeepLabCut: markerless pose esƟmaƟon of user-defined body parts with deep learning. Nat Neurosci 21:1281– 1289. Michaels JA, Dann B, Scherberger H (2016) Neural PopulaƟon Dynamics during Reaching Are BeƩ er Explained by a Dynamical System than RepresentaƟonal Tuning. PLoS Comput Biol 12:e1005175- Available at: hƩ ps://doi.org/10.1371/journal.pcbi.1005175. Montuelle SJ, Olson RA, CurƟs H, Williams SH (2020) Unilateral lingual nerve transecƟon alters jaw- tongue coordinaƟon during masƟcaƟon in pigs. J Appl Physiol (1985) 128:941–951 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/32191597. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Murray GM, Sessle BJ (1992) FuncƟonal properƟes of single neurons in the face primary motor cortex of the primate. III. RelaƟons with different direcƟons of trained tongue protrusion. J Neurophysiol 67:775–785 Available at: hƩ p://dx.doi.org/10.1152/jn.1992.67.3.775. Prud’homme MJ, Kalaska JF (1994) PropriocepƟve acƟvity in primate primary somatosensory cortex during acƟve arm reaching movements. J Neurophysiol 72:2280–2301 Available at: hƩ p://dx.doi.org/10.1152/jn.1994.72.5.2280. Punacha S, Huang K, Arce-McShane FI (2024) Effects of healthy aging on tongue-jaw kinemaƟcs during feeding behavior in rhesus macaques. bioRxiv:2024.07.31.605680 Available at: hƩ p://biorxiv.org/content/early/2024/08/03/2024.07.31.605680.abstract. Rickert J, Riehle A, Aertsen A, RoƩ er S, Nawrot MP (2009) Dynamic Encoding of Movement DirecƟon in Motor CorƟcal Neurons. The Journal of Neuroscience 29:13870 Available at: hƩ p://www.jneurosci.org/content/29/44/13870.abstract. Russo AA, BiƩ ner SR, Perkins SM, Seely JS, London BM, Lara AH, Miri A, Marshall NJ, Kohn A, Jessell TM, AbboƩ LF, Cunningham JP , Churchland MM (2018) Motor Cortex Embeds Muscle-like Commands in an Untangled PopulaƟon Response. Neuron 97:953-966.e8 Available at: hƩ p://dx.doi.org/10.1016/j.neuron.2018.01.004. Santhanam G, Yu BM, Gilja V, Ryu SI, Afshar A, Sahani M, Shenoy K V (2009) Factor-Analysis Methods for Higher-Performance Neural Prostheses. J Neurophysiol 102:1315–1330 Available at: hƩ ps://doi.org/10.1152/jn.00097.2009. Schwartz AB, KeƩ ner RE, Georgopoulos AP (1988a) Primate motor cortex and free arm movements to visual targets in three-dimensional space. I. relaƟons between single cell discharge and direcƟon of movement. JNeurosci 8:2913–2927. Schwartz AB, KeƩ ner RE, Georgopoulos AP (1988b) Primate motor cortex and free arm movements to visual targets in three-dimensional space. I. relaƟons between single cell discharge and direcƟon of movement. JNeurosci 8:2913–2927. Seely JS, Kaufman MT, Ryu SI, Shenoy K V, Cunningham JP , Churchland MM (2016) Tensor Analysis Reveals DisƟnct PopulaƟon Structure that Parallels the Different ComputaƟonal Roles of Areas M1 and V1. PLoS Comput Biol 12:e1005164- Available at: hƩ ps://doi.org/10.1371/journal.pcbi.1005164. Sessle BJ, Yao D, Nishiura H, Yoshino K, Lee J, MarƟn RE, Murray GM (2005a) ProperƟes and plasƟcity of the primate somatosensory and motor cortex related to orofacial sensorimotor funcƟon. Clin Exp Pharmacol Physiol 32:109–114 Available at: hƩ p://dx.doi.org/10.1111/j.1440-1681.2005.04137.x. Sessle BJ, Yao D, Nishiura H, Yoshino K, Lee JC, MarƟn RE, Murray GM, Yao B, Nishiura H, Yoshino K, Lee JC, MarƟn RE, Murray GM (2005b) ProperƟes and plasƟcity of the primate somatosensory and motor cortex related to orofacial sensorimotor funcƟon. Clin Exp Pharmacol Physiol 32:109–114 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/15730444. Shinagawa H, Ono T, Ishiwata Y , Honda E, Sasaki T, Taira M, Iriki A, Kuroda T (2003) Hemispheric Dominance of Tongue Control Depends on the Chewing-side Preference. J Dent Res 82:278–283. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Smith JH, Cutrer FM (2011) Numbness maƩ ers: A clinical review of trigeminal neuropathy. Cephalalgia 31:1131–1144 Available at: hƩ p://dx.doi.org/10.1177/0333102411411203. Tabot GA, Dammann JF, Berg JA, Tenore F V, Boback JL, Vogelstein RJ, Bensmaia SJ (2013) Restoring the sense of touch with a prostheƟc hand through a brain interface. Proc Natl Acad Sci U S A 110:18279–18284 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/24127595. Takizawa C, Gemmell E, Kenworthy J, Speyer R (2016) A SystemaƟc Review of the Prevalence of Oropharyngeal Dysphagia in Stroke, Parkinson’s Disease, Alzheimer’s Disease, Head Injury, and Pneumonia. Dysphagia 31:434–441 Available at: hƩ p://dx.doi.org/10.1007/s00455-016- 9695-9. Toda T, Taoka M (2002) Hierarchical somestheƟc processing of tongue inputs in the postcentral somatosensory cortex of conscious macaque monkeys. Exp Brain Res 147:243–251 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/12410339. Toda T, Taoka M (2004) Converging paƩ erns of inputs from oral structures in the postcentral somatosensory cortex of conscious macaque monkeys. Exp Brain Res 158:43–49 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/15014923. Toda T, Taoka M (2006) Postcentral neurons with covert recepƟve fields in conscious macaque monkeys: their selecƟve responsiveness to simultaneous two-point sƟmuli applied to discrete oral porƟons. Exp Brain Res 168:303–306 Available at: hƩ p://www.ncbi.nlm.nih.gov/pubmed/16307237. Todrank J, Bartoshuk LM (1991) A taste illusion: Taste sensaƟon localized by touch. Physiol Behav 50:1027–1031 Available at: hƩ ps://www.sciencedirect.com/science/arƟcle/pii/003193849190432N. Vilela M, Hochberg LR (2020) ApplicaƟons of brain-computer interfaces to the control of roboƟc and prostheƟc arms. In: Handbook of Clinical Neurology, pp 87–99. Xu D, Dong M, Chen Y , Delgado AM, Hughes NC, Zhang L, O’Connor DH (2022) CorƟcal processing of flexible and context-dependent sensorimotor sequences. Nature 603:464–469 Available at: hƩ ps://pubmed.ncbi.nlm.nih.gov/35264793. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 1. DirecƟon of tongue mo Ɵon in each behavioral task. (A) SchemaƟc of the loca Ɵon of three spouts, leŌ (L), middle (M), and right (R), for the drinking task. Tongue direcƟon was categorized based on spout loca Ɵon. (B) CalculaƟon of 3D tongue direc Ɵon during feeding. θ is the instantaneous 3D direcƟon of the tongue Ɵp over a 100 ms interval between its posiƟons at t1 and t2, where t1 = 0 and t2 = t1 + 100. The doƩ ed line shows the actual trajectory during this interval. Figure 2. Examples of single neuron activity in relation to tongue direction. (A) Each peri-event time histogram (PETH and ±1 SE, smoothed by a 25- ms Gaussian kernel) corresponds to spiking activity for a specific range of tongue direction for feeding trials. Dashed lines indicate 100- ms interval used for calculating the tongue direction. (B) PETHs for drinking trials with the same spout, centered at the point of minimum protrusion of the tongue (0-s). Percent tongue displacement along the anterior-posterior axis is shown in grey, with shaded area representing ±1 SD. Vertical lines indicate 500-ms interval used for tuning analysis. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 3. Directional tuning of neurons during control tasks. (A) 3D firing rate map of a neuron in MIo during feeding. Smaller inset plots are 1D tuning curves across each axis. (B) Percentage of neurons tuned to direction, combined for both subjects. Recordings were taken from four areas of the OSMCx: rMIo - rostral M1, cMIo - caudal M1, SIo(3a/3b) - area 3a/3b, and SIo(1/2) - area 1/2. Error bars represent ±1 SE. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 4. DirecƟonal tuning to yaw and pitch during feeding. (A) Firing rate maps of a neuron in MIo and in SIo across yaw and pitch angles. Firing rates were averaged across all 100 ms feeding intervals within a 10° range. (B) Proportion of neurons tuned to yaw and pitch, combined for both subjects. Recordings were taken from four areas of the OSMCx: rM1 - rostral M1, cM1 - caudal M1, SC(3a/3b) - area 3a/3b , and SC(1/2) - area 1/2. Error bars represent ±1 SE. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 5. Cosine tuning of MIo and SIo neurons. (A) DistribuƟon of 3D preferred direcƟons in unit sphere for neurons that fit the tuning funcƟon during feeding, combined for both subjects. The origin represents the start of a movement. Color bar represents posterior-anterior axis. (B) DistribuƟon of the index for the depth of direcƟonal tuning, combined for both subjects. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 6. DistribuƟon of PDs in MIo (yellow) and SIo (purple) neurons during control feeding (A) and drinking (B). For the feeding task, polar plots are split into 10  bins with thick colored lines represenƟng the mean PD. For the drinking task, error bars represent ±1 SE. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 7. Neural populaƟon trajectories vary across direc Ɵons. Trial-averaged trajectories of MIo and SIo populaƟon acƟvity along the first three latent factors for Monkey R, grouped by direc Ɵon. Axes for SIo are 1/4 scale of MIo. Arrows indicate the end of the trajectory. Percentages denote the sum of the variance explained by the first three factors. Inset plots show the difference between the average inter- trajectory distances of MIo and SIo over Ɵme for both feeding and drinking. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 8. Effect of nerve block on direcƟon of tongue movement. (A) DistribuƟon of tongue direcƟons during feeding. (B) Variance in 3D trajectory endpoints during drinking (Posterior-Anterior, Inferior-Superior, Le Ō - Right) for each direc Ɵon: leŌ (L), middle (M), right (R). (C) VariaƟon in the distance of drinking endpoint posiƟons from the mean endpoint. Le Ō halves of hemi-violins (black) are control and right halves (red) are nerve block for an individual. Horizontal black lines represent the mean and horizontal red lines the median.

of two-tailed t-test and f-test are indicated by asterisks and crosses, respecƟvely: *,† p < 0.05; **,†† p < 0.01; ***,††† p < 0.001. Smaller inset plots show that there was no effect in the sham nerve block condiƟon, for reference. The sham procedure was iden Ɵcal to the nerve block, except the anesthe Ɵc was subsƟtuted with saline soluƟon .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 9. Effect of nerve block on drinking kinemaƟcs in Monkey R. (A) Tongue Ɵp trajectories from star Ɵng posi Ɵon to one of three drinking spouts in the control and nerve block condiƟons. (B) Drinking trajectory endpoints, where the black dot represents the mean endpoint posiƟon. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 10. Effects of nerve block on direc Ɵonal tuning of OSMCx neurons during feeding and drinking tasks. (A) Percentage of direcƟonally tuned neurons in four areas: rMIo - rostral M1, cMIo - caudal M1, SIo(3a/3b) - area 3a/3b, and SIo(1/2) - area 1/2 . Filled in bars represent control while empty bars represent nerve block. Error bars represent ±1 SE. (B) Percentage of MIo and SIo neurons which gained or lost direcƟonality with the addiƟon of nerve block. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 11. Effects of nerve block on the distribuƟon of PDs of MIo (yellow) and SIo (purple) neurons. (A) For the feeding task, polar plots are split into 10° bins with thick colored lines represenƟng the mean PD. Significant circular concentra Ɵon test (k- test) comparing control and nerve block are indicated by asterisks: *p < 0.05; **p < 0.01; ***p < 0.001. (B) For the drinking task, error bars represent ±1 SE. Filled in bars represent control while empty bars represent nerve block. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 12. Effect of nerve block on popula Ɵon trajectories. Trial- averaged trajectories of MIo and SIo populaƟon acƟvity for Monkey R’s feeding and drinking sessions, grouped by direcƟon. Axes represent the latent factors from control data, with the x-axis chosen as the factor with the highest degree of separaƟon between direcƟons. Lighter, doƩ ed lines represent superimposed populaƟon trajectories in the nerve block condiƟon. Insets show the difference between the average inter-trajectory distances for control and nerve block condiƟons. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint Figure 13. Accuracies of two different decoding algorithms from MIo and SIo populations of equal size (N=28). (A) Comparison between average decoding accuracy of KNN classifier. Chance level is 33.33%. (B) Comparison between average decoding accuracy by LSTM network. Data shown separately for each subject, behavioral task, and condition. The dashed line signifies equal decoding performance for MIo and SIo. Decoding accuracies from full populations are included in Figure 13 – figure supplement 2. .CC-BY 4.0 International licenseavailable under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (whichthis version posted April 24, 2025. ; https://doi.org/10.1101/2024.07.02.601741doi: bioRxiv preprint